id
stringlengths 20
52
| title
stringlengths 3
459
| abstract
stringlengths 0
12.3k
| classification_labels
list | numerical_classification_labels
list |
|---|---|---|---|---|
SCOPUS_ID:85068267709
|
A Distributed Ensemble of Deep Convolutional Neural Networks with Random Forest for Big Data Sentiment Analysis
|
Big data has become an important issue for a large number of research areas. With the advent of social networks, users can express their feelings about the products they bought or the services they used every day. Also, they can share their ideas and interests, discuss current issues. Therefore, Big Data sentiment analysis has become important in decision-making processes. In this paper, we propose a novel distributed ensemble of deep convolutional neural networks with random forest for sentiment analysis, which is tailored to handle large-scale data and improve classification accuracy. Experimental results on two real-world data sets confirm the claim.
|
[
"Sentiment Analysis"
] |
[
78
] |
SCOPUS_ID:85056312030
|
A Distributed Representation Model for Short Text Analysis
|
The distributed representation of short texts has become an important task in text mining. However, the direct application of the traditional Paragraph Vector may not be suitable, and the fundamental reason is that it does not make use of the information of corpus in training process, so it can not effectively improve the situation of insufficient contextual information in short texts. In view of this, in this paper we propose a novel distributed representation model for short texts called BTPV (biterm topic paragraph vector). BTPV adds the topic information of BTM (biterm topic model) to the Paragraph Vector model. This method not only uses the global information of corpus, but also perfects the implicit vector of Paragraph Vector with the explicit topic information of BTM. At last, we crawl popular news comments from the Internet as experimental data sets, using K-Means clustering algorithm to compare the models' representation performance. Experimental results have shown that the BTPV model can get better clustering results compared with the common distributed representation models such as word2vec and Paragraph Vector, which indicates the advantage of the proposed model for short text analysis.
|
[
"Topic Modeling",
"Semantic Text Processing",
"Representation Learning",
"Text Clustering",
"Information Extraction & Text Mining"
] |
[
9,
72,
12,
29,
3
] |
SCOPUS_ID:85055421962
|
A Distributed Text Clustering Model Based on Multi-Agent
|
As the Internet big data grow rapidly, it urgently needs us to design new clustering approaches that can handle large-scale semi-structured and unstructured text data. The existing methods have the following disadvantages: the commonly used text datasets are very monotonous, the accuracy of text clustering on semi-structured and unstructured Web texts is very low, and the efficiency of clustering can't be guaranteed when the cardinality of documents is very large. Aiming to cope with these drawbacks in existing methods, a new clustering model based on swarm intelligence was proposed, called Switch (a Swarm intelligence based text clustering algorithm), which can support multiple languages including Tibetan, Chinese, and English as well. The basic idea of the proposed method is that: it first constructs the vector space model and then obtains the feature vector set of texts by employing the natural language processing and data preprocessing techniques. The parameters of the proposed swarm intelligence based text clustering algorithm are initialized, and the agents can randomly move in a two dimensional text space. The agents calculate the similarity of texts in the grids where they currently stay in to other texts, and use the probability transition function to calculate the probability of picking up and dropping down texts. A distributed dynamic text stream clustering architecture based on multi-agent was proposed, and the proposed distributed architecture was applied to the swarm intelligence based text clustering approach. The distributed working environment of swarm intelligence is designed to be a set of soft agents through communication. Three agents were proposed, including similarity calculation agents, state awareness agents and text parsing agents. By coping with the problems of agent states synchronization, the cost of communication between processors, and load balancing of processors, the calculation tasks are partitioned into different subtasks and the processors perform these tasks in a distributed fashion. In addition, the working mechanism of the proposed distributed swarm intelligent clustering approach based on multi-agent was introduced and the distributed communication schema was given, by which the agents can communicate with others and collaborate with each other to complete the task of text clustering. The distributed clustering on computer clusters can be achieved by the middleware of JADE based on multi-agents, and its advantages include: it has better distributed computing power and large memory processing capability than the stand-alone processing, and employs JADE middleware to perform communication and cooperation among agents in order to complete text clustering efficiently. Experiments were conducted on real semi-structured Web text datasets including Tibetan, Chinese and English. By taking Tibetan as an example, the results show that: the clustering accuracy of the proposed distributed clustering approach is averagely improved by 12.2% and 3.8% and the time cost is reduced by 73.0% and 50.6% on average by comparing to the k-means and stand-alone single node cluster. The results show that when the number of agents is between 150 and 250 in the computer cluster with n nodes, the time cost of text clustering might approximate to 1/n time cost with regard to a stand-alone node.
|
[
"Responsible & Trustworthy NLP",
"Text Clustering",
"Information Extraction & Text Mining",
"Green & Sustainable NLP"
] |
[
4,
29,
3,
68
] |
SCOPUS_ID:85077131787
|
A Distributed Topic Model for Large-Scale Streaming Text
|
Learning topic information from large-scale unstructured text has attracted extensive attention from both the academia and industry. Topic models, such as LDA and its variants, are a popular machine learning technique to discover such latent structure. Among them, online variational hierarchical Dirichlet process (onlineHDP) is a promising candidate for dynamically processing streaming text. Instead of a static assignment in advance, the number of topics in onlineHDP is inferred from the corpus as the training process proceeds. However, when dealing with large scale streaming data it still suffers from the limited model capacity problem. To this end, we proposed a distributed version of the onlineHDP algorithm (named as DistHDP) in this paper, the training task is split into many sub-batch tasks and distributed across multiple worker nodes, such that the whole training process is accelerated. The model convergence is guaranteed through a distributed variation inference algorithm. Extensive experiments conducted on several real-world datasets demonstrate the usability and scalability of the proposed algorithm.
|
[
"Topic Modeling",
"Information Extraction & Text Mining"
] |
[
9,
3
] |
http://arxiv.org/abs/2012.11635v2
|
A Distributional Approach to Controlled Text Generation
|
We propose a Distributional Approach for addressing Controlled Text Generation from pre-trained Language Models (LMs). This approach permits to specify, in a single formal framework, both "pointwise" and "distributional" constraints over the target LM -- to our knowledge, the first model with such generality -- while minimizing KL divergence from the initial LM distribution. The optimal target distribution is then uniquely determined as an explicit EBM (Energy-Based Model) representation. From that optimal representation we then train a target controlled Autoregressive LM through an adaptive distributional variant of Policy Gradient. We conduct a first set of experiments over pointwise constraints showing the advantages of our approach over a set of baselines, in terms of obtaining a controlled LM balancing constraint satisfaction with divergence from the initial LM. We then perform experiments over distributional constraints, a unique feature of our approach, demonstrating its potential as a remedy to the problem of Bias in Language Models. Through an ablation study, we show the effectiveness of our adaptive technique for obtaining faster convergence. (Code available at https://github.com/naver/gdc)
|
[
"Language Models",
"Semantic Text Processing",
"Text Generation"
] |
[
52,
72,
47
] |
http://arxiv.org/abs/2210.02889v2
|
A Distributional Lens for Multi-Aspect Controllable Text Generation
|
Multi-aspect controllable text generation is a more challenging and practical task than single-aspect control. Existing methods achieve complex multi-aspect control by fusing multiple controllers learned from single-aspect, but suffer from attribute degeneration caused by the mutual interference of these controllers. To address this, we provide observations on attribute fusion from a distributional perspective and propose to directly search for the intersection areas of multiple attribute distributions as their combination for generation. Our method first estimates the attribute space with an autoencoder structure. Afterward, we iteratively approach the intersections by jointly minimizing distances to points representing different attributes. Finally, we map them to attribute-relevant sentences with a prefix-tuning-based decoder. Experiments on the three-aspect control task, including sentiment, topic, and detoxification aspects, reveal that our method outperforms several strong baselines on attribute relevance and text quality and achieves the SOTA. Further analysis also supplies some explanatory support for the effectiveness of our approach.
|
[
"Text Generation"
] |
[
47
] |
http://arxiv.org/abs/2106.15772v1
|
A Diverse Corpus for Evaluating and Developing English Math Word Problem Solvers
|
We present ASDiv (Academia Sinica Diverse MWP Dataset), a diverse (in terms of both language patterns and problem types) English math word problem (MWP) corpus for evaluating the capability of various MWP solvers. Existing MWP corpora for studying AI progress remain limited either in language usage patterns or in problem types. We thus present a new English MWP corpus with 2,305 MWPs that cover more text patterns and most problem types taught in elementary school. Each MWP is annotated with its problem type and grade level (for indicating the level of difficulty). Furthermore, we propose a metric to measure the lexicon usage diversity of a given MWP corpus, and demonstrate that ASDiv is more diverse than existing corpora. Experiments show that our proposed corpus reflects the true capability of MWP solvers more faithfully.
|
[
"Reasoning",
"Numerical Reasoning"
] |
[
8,
5
] |
http://arxiv.org/abs/2109.11834v1
|
A Diversity-Enhanced and Constraints-Relaxed Augmentation for Low-Resource Classification
|
Data augmentation (DA) aims to generate constrained and diversified data to improve classifiers in Low-Resource Classification (LRC). Previous studies mostly use a fine-tuned Language Model (LM) to strengthen the constraints but ignore the fact that the potential of diversity could improve the effectiveness of generated data. In LRC, strong constraints but weak diversity in DA result in the poor generalization ability of classifiers. To address this dilemma, we propose a {D}iversity-{E}nhanced and {C}onstraints-\{R}elaxed {A}ugmentation (DECRA). Our DECRA has two essential components on top of a transformer-based backbone model. 1) A k-beta augmentation, an essential component of DECRA, is proposed to enhance the diversity in generating constrained data. It expands the changing scope and improves the degree of complexity of the generated data. 2) A masked language model loss, instead of fine-tuning, is used as a regularization. It relaxes constraints so that the classifier can be trained with more scattered generated data. The combination of these two components generates data that can reach or approach category boundaries and hence help the classifier generalize better. We evaluate our DECRA on three public benchmark datasets under low-resource settings. Extensive experiments demonstrate that our DECRA outperforms state-of-the-art approaches by 3.8% in the overall score.
|
[
"Low-Resource NLP",
"Language Models",
"Semantic Text Processing",
"Information Retrieval",
"Information Extraction & Text Mining",
"Text Classification",
"Responsible & Trustworthy NLP"
] |
[
80,
52,
72,
24,
3,
36,
4
] |
SCOPUS_ID:85097335169
|
A Divide-and-Conquer Approach to the Summarization of Long Documents
|
We present a novel divide-and-conquer method for the neural summarization of long documents. Our method exploits the discourse structure of the document and uses sentence similarity to split the problem into an ensemble of smaller summarization problems. In particular, we break a long document and its summary into multiple source-target pairs, which are used for training a model that learns to summarize each part of the document separately. These partial summaries are then combined in order to produce a final complete summary. With this approach we can decompose the problem of long document summarization into smaller and simpler problems, reducing computational complexity and creating more training examples, which at the same time contain less noise in the target summaries compared to the standard approach. We demonstrate that this approach paired with different summarization models, including sequence-to-sequence RNNs and Transformers, can lead to improved summarization performance. Our best models achieve results that are on par with the state-of-the-art in two two publicly available datasets of academic articles.
|
[
"Summarization",
"Text Generation",
"Information Extraction & Text Mining"
] |
[
30,
47,
3
] |
SCOPUS_ID:85101969114
|
A Document Clustering Approach Using Shared Nearest Neighbour Affinity, TF-IDF and Angular Similarity
|
Quantum of data is increasing in an exponential order. Clustering is a major task in many text mining applications. Organizing text documents automatically, extracting topics from documents, retrieval of information and information filtering are considered as the applications of clustering. This task reveals identical patterns from a collection of documents. Understanding of the documents, representation of them and categorization of documents require various techniques. Text clustering process requires both natural language processing and machine learning techniques. An unsupervised spatial pattern identification approach is proposed for text data. A new algorithm for finding coherent patterns from a huge collection of text data is proposed, which is based on the shared nearest neighbour. The implementation followed by validation confirms that the proposed algorithm can cluster the text data for the identification of coherent patterns. The results are visualized using a graph. The results show the methodology works well for different text datasets.
|
[
"Information Extraction & Text Mining",
"Information Retrieval",
"Text Classification",
"Text Clustering"
] |
[
3,
24,
36,
29
] |
SCOPUS_ID:85075765279
|
A Document Driven Dialogue Generation Model
|
Most of the current man-machine dialogues are at the two end-points of a spectrum of dialogues, i.e. goal-driven dialogues and non goal-driven chit-chats. Document-driven dialogues provide a bridge between them with the change of documents from structured data to unstructured free texts. This paper proposes a Document Driven Dialogue Generation model (D3G) which generates dialogues centering a given document, as well as answering user’s questions. A Doc-Reader mechanism is designed to locate the content related to user’s questions in documents. A Multi-Copy mechanism is employed to generate document-related responses. And the dialogue history is used in both mechanisms. Experimental results on the CMU_DOG dataset show that our D3G model can not only generate informative responses that are more relevant to the document, but also answer user’s questions better than the baseline models.
|
[
"Dialogue Response Generation",
"Natural Language Interfaces",
"Text Generation",
"Dialogue Systems & Conversational Agents"
] |
[
14,
11,
47,
38
] |
SCOPUS_ID:85147841547
|
A Document Image Quality Assessment Algorithm Based on Information Entropy in Text Region
|
The quality of the image is critical to Optical Character Recognition (OCR), poor quality images will lead OCR to generate unreliable results. There are relative high ratio of low quality images in practical OCR-based application scenarios, how to evaluate quality of image and filter out unqualified images by document image quality assessment (DIQA) algorithms effectively is a big challenge for these scenarios. Current DIQA algorithms mainly focus on the overall image features rather than the text region, while the quality of the text region is dominant factor for OCR. In this paper, we propose a document image quality assessment algorithm based on information entropy in text region of image. Our algorithmic framework mainly consists of three networks to detect, extract and evaluate text region in image respectively. We build a quality prediction network based on HyperNet, and use the information entropy of the text region as the score weight, so that the final score can reflect the quality of the text region better. Finally, testing results on benchmark dataset SmartDoc-QA and our constructed dataset DocImage1k demonstrate that the proposed algorithm achieves excellent performance.
|
[
"Visual Data in NLP",
"Multimodality"
] |
[
20,
74
] |
SCOPUS_ID:0016049546
|
A Document Storage Method Based on Polarized Distance
|
Some elementary mathematical properties of term matching document retrieval systems are developed. These properties are used as a basis for a new file organization technique. Some of the advantages of this new method are (1) the key-to-address transformation is easily determined; (2) the documentary information is stored only once in the file; (3) the file organization allows the use of various matching functions and thresholds; and (4) the dimensionality of the transform is easily expanded to accommodate various sized data bases. © 1974, ACM. All rights reserved.
|
[
"Document Retrieval",
"Information Retrieval"
] |
[
56,
24
] |
SCOPUS_ID:85119320867
|
A Document-Level Machine Translation Quality Estimation Model Based on Centering Theory
|
Machine translation Quality Estimation (QE) aims to estimate the quality of machine translations without relying on golden references. Current QE researches mainly focus on sentence-level QE models, which could not capture discourse-related translation errors. To tackle this problem, this paper presents a novel document-level QE model based on Centering Theory (CT), which is a linguistics theory for assessing discourse coherence. Furthermore, we construct and release an open-source Chinese-English corpus at https://github.com/ydc/cpqe for document-level machine translation QE, which could be used to support further studies. Finally, experimental results show that the proposed model significantly outperformed the baseline model.
|
[
"Machine Translation",
"Linguistic Theories",
"Text Generation",
"Linguistics & Cognitive NLP",
"Multilinguality"
] |
[
51,
57,
47,
48,
0
] |
http://arxiv.org/abs/1906.04362v1
|
A Document-grounded Matching Network for Response Selection in Retrieval-based Chatbots
|
We present a document-grounded matching network (DGMN) for response selection that can power a knowledge-aware retrieval-based chatbot system. The challenges of building such a model lie in how to ground conversation contexts with background documents and how to recognize important information in the documents for matching. To overcome the challenges, DGMN fuses information in a document and a context into representations of each other, and dynamically determines if grounding is necessary and importance of different parts of the document and the context through hierarchical interaction with a response at the matching step. Empirical studies on two public data sets indicate that DGMN can significantly improve upon state-of-the-art methods and at the same time enjoys good interpretability.
|
[
"Natural Language Interfaces",
"Information Retrieval",
"Dialogue Systems & Conversational Agents"
] |
[
11,
24,
38
] |
SCOPUS_ID:85137353354
|
A Dog Is Passing over the Jet? A Text-Generation Dataset for Korean Commonsense Reasoning and Evaluation
|
Recent natural language understanding (NLU) research on the Korean language has been vigorously maturing with the advancements of pretrained language models and datasets. However, Korean pretrained language models still struggle to generate a short sentence with a given condition based on compositionality and commonsense reasoning (i.e., generative commonsense reasoning). The two major challenges are inadequate data resources to develop generative commonsense reasoning regarding Korean linguistic features and to evaluate language models which are necessary for natural language generation (NLG). To solve these problems, we propose a textgeneration dataset for Korean generative commonsense reasoning and language model evaluation. In this work, a semi-automatic dataset construction approach filters out contents inexplicable to commonsense, ascertains quality, and reduces the cost of building the dataset. We also present an in-depth analysis of the generation results of language models with various evaluation metrics along with human-annotated scores. The whole dataset is publicly available at (https://aihub.or. kr/opendata/korea-university).
|
[
"Commonsense Reasoning",
"Language Models",
"Reasoning",
"Semantic Text Processing"
] |
[
62,
52,
8,
72
] |
https://aclanthology.org//W15-4703/
|
A Domain Agnostic Approach to Verbalizing n-ary Events without Parallel Corpora
|
[
"Text Generation"
] |
[
47
] |
|
SCOPUS_ID:85146913076
|
A Domain Specific Parallel Corpus and Enhanced English-Assamese Neural Machine Translation
|
Machine translation deals with automatic translation from one natural language to another. Neural machine translation is a widely accepted technique of the corpus-based machine translation approach. However, an adequate amount of training data is required, and there is a need for the domain-wise parallel corpus to improve translational performance that shows translational coverages in various domains. In this work, a domain-specific parallel corpus is prepared that includes different domain coverages, namely, Agriculture, Government Office, Judiciary, Social Media, Tourism, COVID-19, Sports, and Literature domains for low-resource English-Assamese pair translation. Moreover, we have tackled data scarcity and word-order divergence problems via data augmentation and prior alignment concept. Also, we have contributed Assamese pretrained LM, Assamese word-embeddings by utilizing Assamese monolingual data, and a bilingual dictionary-based post-processing step to enhance transformer-based neural machine translation. We have achieved state-of-the-art results for both forward (English-to-Assamese) and backward (Assamese-to-English) directions of translation.
|
[
"Multilinguality",
"Language Models",
"Low-Resource NLP",
"Machine Translation",
"Semantic Text Processing",
"Text Generation",
"Responsible & Trustworthy NLP"
] |
[
0,
52,
80,
51,
72,
47,
4
] |
SCOPUS_ID:84902377604
|
A Domain independent double layered approach to keyphrase generation
|
The annotation of documents and web pages with semantic metatdata is an activity that can greatly increase the accuracy of Information Retrieval and Personalization systems, but the growing amount of text data available is too large for an extensive manual process. On the other hand, automatic keyphrase generation, a complex task involving Natural Language Processing and Knowledge Engineering, can significantly support this activity. Several different strategies have been proposed over the years, but most of them require extensive training data, which are not always available, suffer high ambiguity and differences in writing style, are highly domainspecific, and often rely on a well-structured knowledge that is very hard to acquire and encode. In order to overcome these limitations, we propose in this paper an innovative domain-independent approach that consists of an unsupervised keyphrase extraction phase and a subsequent keyphrase inference phase based on loosely structured, collaborative knowledge such as Wikipedia, Wordnik, and Urban Dictionary. This double layered approach allows us to generate keyphrases that both describe and classify the text.
|
[
"Term Extraction",
"Text Generation",
"Information Extraction & Text Mining"
] |
[
1,
47,
3
] |
SCOPUS_ID:85105253092
|
A Double Channel CNN-LSTM Model for Text Classification
|
The CNN-LSTM model has the advantages of combining Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM). It can perform timing analysis while extracting abstract features. It is widely used in Computer Vision and Natural Language Processing (NLP) fields and has achieved satisfactory results. However, for a large number of samples of complex text data, especially for words with ambiguous meanings, the word-level CNN-LSTM model is insufficient. Therefore, in order to solve this issue, this paper presents an improved Double Channel (DC) mechanism as a significant enhancement to CNN-LSTM. In this DC mechanism, two channels are used to receive word-level and char-level embedding respectively, at the same time. Hybrid Attention is proposed to combine the current time output with the current time unit state, and then using Attention to calculate the weight. By calculating the probability distribution of each timestep input data weight, the weight score is obtained, and then weighted summation is performed, and the data input by each timestep is subjected to trade-off learning to improve the generalization ability of the model learning. After experimental comparison, the DC CNN-LSTM model proposed in this paper has significantly superior accuracy and F1-score compared with the basic CNN-LSTM model.
|
[
"Language Models",
"Semantic Text Processing",
"Text Classification",
"Information Retrieval",
"Information Extraction & Text Mining"
] |
[
52,
72,
36,
24,
3
] |
SCOPUS_ID:85143495464
|
A Double Meta<sup>n</sup>-Semantic Search Model Based on Ontology and Semantic Similarity: Asthma Disease
|
With the exponential and rapid growth of online resources in recent years, there has been a huge increase in the use of search engines; these are also one of the most common ways to navigate the Web content without taking into account, in general, the request meaning by which was successfully added the user's webpage provides us with a lot of results. This problem has led to the integration of semantics in the search for information on the Web (Semantic Web). The use of semantic tools, such as ontology, WordNet dictionary, semantic similarity measure, etc., has contributed to the semantic search development and more particularly, semantic Metan-search. The success of semantic search is closely linked to the availability of domain ontologies. The objective of this paper is to propose a double model of repetitive semantic search, called Double Metan-Semantic Search Model (2∞n-SSM). On the one hand, it is assisted and based on the concepts extracted from the user's search domain ontology, which will permit the user to choose a concept from this list of concepts and launch their search; on the other hand, it is free, in that the user enters their own concept and launches their search. This is based on WordNet tool, user's same search domain ontology and the semantic similarity calculation techniques between concepts in the same ontology. The result of this model is a set of URL links. The term Metan indicates that the search is done in depth (∞n-SS) via choosing each time a URL result by the user. Its experimentation in the asthma disease field gave very promising results in quantity and quality of information via the URL link results (semantic support).
|
[
"Semantic Text Processing",
"Semantic Similarity",
"Knowledge Representation",
"Semantic Search",
"Information Retrieval"
] |
[
72,
53,
18,
41,
24
] |
SCOPUS_ID:33847333264
|
A Double Metaphone encoding for Bangla and its application in spelling checker
|
We present a Double Metaphone encoding for Bangla that can be used by spelling checkers to improve the quality of suggestions for misspelled words. The complex rules of Bangla spelling present a significant challenge in producing suggestions for a misspelled word when employing the traditional edit-distance methods; one must take phonetic similarity into account for the suggested alternatives to be reasonably accurate. We propose a Double Metaphone encoding for Bangla, taking into account the various contest-sensitive rules, including those involving the large repertoire of consonant clusters in Bangla, and present a comparison with the traditional edit-distance based methods in producing suggestions for misspelled words. © 2005 IEEE.
|
[
"Phonetics",
"Syntactic Text Processing"
] |
[
64,
15
] |
http://arxiv.org/abs/2206.09158v1
|
A Double-Graph Based Framework for Frame Semantic Parsing
|
Frame semantic parsing is a fundamental NLP task, which consists of three subtasks: frame identification, argument identification and role classification. Most previous studies tend to neglect relations between different subtasks and arguments and pay little attention to ontological frame knowledge defined in FrameNet. In this paper, we propose a Knowledge-guided Incremental semantic parser with Double-graph (KID). We first introduce Frame Knowledge Graph (FKG), a heterogeneous graph containing both frames and FEs (Frame Elements) built on the frame knowledge so that we can derive knowledge-enhanced representations for frames and FEs. Besides, we propose Frame Semantic Graph (FSG) to represent frame semantic structures extracted from the text with graph structures. In this way, we can transform frame semantic parsing into an incremental graph construction problem to strengthen interactions between subtasks and relations between arguments. Our experiments show that KID outperforms the previous state-of-the-art method by up to 1.7 F1-score on two FrameNet datasets. Our code is availavle at https://github.com/PKUnlp-icler/KID.
|
[
"Semantic Parsing",
"Structured Data in NLP",
"Semantic Text Processing",
"Multimodality"
] |
[
40,
50,
72,
74
] |
SCOPUS_ID:85127519559
|
A Drift-Sensitive Distributed LSTM Method for Short Text Stream Classification
|
Real-world applications especially in the fields of social media have produced massive short text streams. Unlike traditional normal texts, these data present the characteristics of short length, high-volume, high-velocity and variable data distribution etc, which lead to the issues of data sparsity and concept drift. It is hence very challenging for existing short text classification algorithms. Therefore, we propose a flexible Long Short-Term Memory (LSTM) ensemble network based short text stream classification approach, which is implemented in a distributed mode while maintaining the high-accuracy advantage of deep learning models. More specifically, external resource based short text embedding using a pretrained embedding model and CNN is first proposed for the solution to the data sparsity of short texts. Second, to adapt to the high-volume and high-velocity short text streams, a flexible LSTM network is developed and implemented in a distributed mode for classifying short text data streams. Third, a concept drift factor is introduced for adapting to the concept drifts caused by the changing of data distributions. Finally, experiments conducted on three real short text data sets demonstrate that as compared with several state-of-the-art short text (stream) classification approaches, the proposed approach can classify short text streams effectively and efficiently while adapting to concept drifts.
|
[
"Language Models",
"Semantic Text Processing",
"Text Classification",
"Representation Learning",
"Information Retrieval",
"Information Extraction & Text Mining"
] |
[
52,
72,
36,
12,
24,
3
] |
SCOPUS_ID:85141898622
|
A Dual Attention Encoder-Decoder Text Summarization Model
|
A worthy text summarization should represent the fundamental content of the document. Recent studies on computerized text summarization tried to present solutions to this challenging problem. Attention models are employed extensively in text summarization process. Classical attention techniques are utilized to acquire the context data in the decoding phase. Nevertheless, without real and efficient feature extraction, the produced summary may diverge from the core topic. In this article, we present an encoder-decoder attention system employing dual attention mechanism. In the dual attention mechanism, the attention algorithm gathers main data from the encoder side. In the dual attention model, the system can capture and produce more rational main content. The merging of the two attention phases produces precise and rational text summaries. The enhanced attention mechanism gives high score to text repetition to increase phrase score. It also captures the relationship between phrases and the title giving them higher score. We assessed our proposed model with or without significance optimization using ablation procedure. Our model with significance optimization achieved the highest performance of 96.7% precision and the least CPU time among other models in both training and sentence extraction.
|
[
"Language Models",
"Semantic Text Processing",
"Summarization",
"Text Generation",
"Information Extraction & Text Mining"
] |
[
52,
72,
30,
47,
3
] |
SCOPUS_ID:85135396708
|
A Dual Knowledge Aggregation Network for Cross-Domain Sentiment Analysis
|
Cross-domain sentiment analysis (CDSA) is an essential subtask of sentiment analysis. It aims to utilize rich source domain data to conquer the data-hungry problem on target domain. Most existing approaches depending on deep learning mainly concentrate on common features or pivots. However, few of them consider the effect of external Knowledge Graph (KG). In this paper, we propose a Dual Knowledge Aggregation Network for Cross-Domain Sentiment Analysis (DKAN), which leverages prior knowledge from two external KGs. Specifically, DKAN comprises two main parts. One is extracting sentence representation features. The other aims to introduce external knowledge better. Also, we use SenticNet to avoid noise from KG by selecting top-n words and inserting special tokens in sentences. We also conduct empirical analyses on the effectiveness of our model on the Amazon reviews dataset. DKAN achieves promising performance compared with other methods.
|
[
"Semantic Text Processing",
"Structured Data in NLP",
"Knowledge Representation",
"Sentiment Analysis",
"Multimodality"
] |
[
72,
50,
18,
78,
74
] |
SCOPUS_ID:85061447293
|
A Dual Prediction Network for Image Captioning
|
General captioning practice involves a single forward prediction, with the aim of predicting the word in the next timestep given the word in the current timestep. In this paper, we present a novel captioning framework, namely Dual Prediction Network (DPN), which is end-to-end trainable and addresses the captioning problem with dual predictions. Specifically, the dual predictions consist of a forward prediction to generate the next word from the current input word, as well as a backward prediction to reconstruct the input word using the predicted word. DPN has two appealing properties: 1) By introducing an extra supervision signal on the prediction, DPN can better capture the interplay between the input and the target; 2) Utilizing the reconstructed input, DPN can make another new prediction. During the test phase, we average both predictions to formulate the final target sentence. Experimental results on the MS COCO dataset demonstrate that, benefiting from the reconstruction step, both generated predictions in DPN outperform the predictions of methods based on the general captioning practice (single forward prediction), and averaging them can bring a further accuracy boost. Overall, DPN achieves competitive results with state-of-the-art approaches, across multiple evaluation metrics.
|
[
"Visual Data in NLP",
"Captioning",
"Text Generation",
"Multimodality"
] |
[
20,
39,
47,
74
] |
http://arxiv.org/abs/2201.05780v3
|
A Dual Prompt Learning Framework for Few-Shot Dialogue State Tracking
|
Dialogue state tracking (DST) module is an important component for task-oriented dialog systems to understand users' goals and needs. Collecting dialogue state labels including slots and values can be costly, especially with the wide application of dialogue systems in more and more new-rising domains. In this paper, we focus on how to utilize the language understanding and generation ability of pre-trained language models for DST. We design a dual prompt learning framework for few-shot DST. Specifically, we consider the learning of slot generation and value generation as dual tasks, and two prompts are designed based on such a dual structure to incorporate task-related knowledge of these two tasks respectively. In this way, the DST task can be formulated as a language modeling task efficiently under few-shot settings. Experimental results on two task-oriented dialogue datasets show that the proposed method not only outperforms existing state-of-the-art few-shot methods, but also can generate unseen slots. It indicates that DST-related knowledge can be probed from PLM and utilized to address low-resource DST efficiently with the help of prompt learning.
|
[
"Low-Resource NLP",
"Language Models",
"Semantic Text Processing",
"Green & Sustainable NLP",
"Natural Language Interfaces",
"Dialogue Systems & Conversational Agents",
"Responsible & Trustworthy NLP"
] |
[
80,
52,
72,
68,
11,
38,
4
] |
http://arxiv.org/abs/1905.10060v1
|
A Dual Reinforcement Learning Framework for Unsupervised Text Style Transfer
|
Unsupervised text style transfer aims to transfer the underlying style of text but keep its main content unchanged without parallel data. Most existing methods typically follow two steps: first separating the content from the original style, and then fusing the content with the desired style. However, the separation in the first step is challenging because the content and style interact in subtle ways in natural language. Therefore, in this paper, we propose a dual reinforcement learning framework to directly transfer the style of the text via a one-step mapping model, without any separation of content and style. Specifically, we consider the learning of the source-to-target and target-to-source mappings as a dual task, and two rewards are designed based on such a dual structure to reflect the style accuracy and content preservation, respectively. In this way, the two one-step mapping models can be trained via reinforcement learning, without any use of parallel data. Automatic evaluations show that our model outperforms the state-of-the-art systems by a large margin, especially with more than 8 BLEU points improvement averaged on two benchmark datasets. Human evaluations also validate the effectiveness of our model in terms of style accuracy, content preservation and fluency. Our code and data, including outputs of all baselines and our model are available at https://github.com/luofuli/DualLanST.
|
[
"Low-Resource NLP",
"Responsible & Trustworthy NLP",
"Text Generation",
"Text Style Transfer"
] |
[
80,
4,
47,
35
] |
SCOPUS_ID:85121902110
|
A Dual Reinforcement Network for Classical and Modern Chinese Text Style Transfer
|
Text style transfer aims to change the stylistic features of a sentence while preserving its content. Although remarkable progress have been achieved in English style transfer, Chinese style transfer, such as classical and modern Chinese style transfer, still relies heavily on manual process. In this paper, we first construct an unsupervised dual reinforcement model to transfer text between classical and modern Chinese styles using a non-disentangled approach, in which the style-transfer-accuracy reward and content-preservation reward are specially designed for model optimization. Meanwhile, we leverage a priori knowledge-based synonym dictionary to build a pseudo-parallel corpus for pre-training to provide a warm start. Experimental evaluations show that our model outperforms state-of-art networks by a large margin.
|
[
"Text Style Transfer",
"Text Generation"
] |
[
35,
47
] |
SCOPUS_ID:85125186811
|
A Dual Self-Attention based Network for Image Captioning
|
Image captioning technology has become an important solution for intelligent robots to understand image content. How to extract image information effectively is the key to generate accurate and reliable captions. In this paper, we propose a dual self-attention based network (DSAN) for image captioning. Specifically, we design a Dual Self-Attention Module (DSAM) embedded into an encoding-decoding architecture to capture the contextual information in the image, which can adaptively integrate local features with global dependencies. The DSAM can significantly improve the caption results by modeling rich contextual dependencies over local features. Experimental results on the MS COCO dataset show that the proposed DSAN can achieve better performance than existing methods.
|
[
"Visual Data in NLP",
"Language Models",
"Semantic Text Processing",
"Captioning",
"Text Generation",
"Multimodality"
] |
[
20,
52,
72,
39,
47,
74
] |
SCOPUS_ID:85106058010
|
A Dual Simple Recurrent Network Model for Chunking and Abstract Processes in Sequence Learning
|
Although many studies have provided evidence that abstract knowledge can be acquired in artificial grammar learning, it remains unclear how abstract knowledge can be attained in sequence learning. To address this issue, we proposed a dual simple recurrent network (DSRN) model that includes a surface SRN encoding and predicting the surface properties of stimuli and an abstract SRN encoding and predicting the abstract properties of stimuli. The results of Simulations 1 and 2 showed that the DSRN model can account for learning effects in the serial reaction time (SRT) task under different conditions, and the manipulation of the contribution weight of each SRN accounted for the contribution of conscious and unconscious processes in inclusion and exclusion tests in previous studies. The results of human performance in Simulation 3 provided further evidence that people can implicitly learn both chunking and abstract knowledge in sequence learning, and the results of Simulation 3 confirmed that the DSRN model can account for how people implicitly acquire the two types of knowledge in sequence learning. These findings extend the learning ability of the SRN model and help understand how different types of knowledge can be acquired implicitly in sequence learning.
|
[
"Syntactic Text Processing",
"Chunking"
] |
[
15,
43
] |
http://arxiv.org/abs/1810.09154v3
|
A Dual-Attention Hierarchical Recurrent Neural Network for Dialogue Act Classification
|
Recognising dialogue acts (DA) is important for many natural language processing tasks such as dialogue generation and intention recognition. In this paper, we propose a dual-attention hierarchical recurrent neural network for DA classification. Our model is partially inspired by the observation that conversational utterances are normally associated with both a DA and a topic, where the former captures the social act and the latter describes the subject matter. However, such a dependency between DAs and topics has not been utilised by most existing systems for DA classification. With a novel dual task-specific attention mechanism, our model is able, for utterances, to capture information about both DAs and topics, as well as information about the interactions between them. Experimental results show that by modelling topic as an auxiliary task, our model can significantly improve DA classification, yielding better or comparable performance to the state-of-the-art method on three public datasets.
|
[
"Text Classification",
"Natural Language Interfaces",
"Dialogue Systems & Conversational Agents",
"Information Retrieval",
"Information Extraction & Text Mining"
] |
[
36,
11,
38,
24,
3
] |
SCOPUS_ID:85118193883
|
A Dual-Attention Neural Network for Pun Location and Using Pun-Gloss Pairs for Interpretation
|
Pun location is to identify the punning word (usually a word or a phrase that makes the text ambiguous) in a given short text, and pun interpretation is to find out two different meanings of the punning word. Most previous studies adopt limited word senses obtained by WSD(Word Sense Disambiguation) technique or pronunciation information in isolation to address pun location. For the task of pun interpretation, related work pays attention to various WSD algorithms. In this paper, a model called DANN (Dual-Attentive Neural Network) is proposed for pun location, effectively integrates word senses and pronunciation with context information to address two kinds of pun at the same time. Furthermore, we treat pun interpretation as a classification task and construct pun-gloss pairs as processing data to solve this task. Experiments on the two benchmark datasets show that our proposed methods achieve new state-of-the-art results. Our source code is available in the public code repository (https://github.com/LawsonAbs/pun ).
|
[
"Explainability & Interpretability in NLP",
"Semantic Text Processing",
"Word Sense Disambiguation",
"Responsible & Trustworthy NLP"
] |
[
81,
72,
65,
4
] |
http://arxiv.org/abs/2109.03587v2
|
A Dual-Channel Framework for Sarcasm Recognition by Detecting Sentiment Conflict
|
Sarcasm employs ambivalence, where one says something positive but actually means negative, and vice versa. The essence of sarcasm, which is also a sufficient and necessary condition, is the conflict between literal and implied sentiments expressed in one sentence. However, it is difficult to recognize such sentiment conflict because the sentiments are mixed or even implicit. As a result, the recognition of sophisticated and obscure sentiment brings in a great challenge to sarcasm detection. In this paper, we propose a Dual-Channel Framework by modeling both literal and implied sentiments separately. Based on this dual-channel framework, we design the Dual-Channel Network~(DC-Net) to recognize sentiment conflict. Experiments on political debates (i.e. IAC-V1 and IAC-V2) and Twitter datasets show that our proposed DC-Net achieves state-of-the-art performance on sarcasm recognition. Our code is released to support research.
|
[
"Stylistic Analysis",
"Sentiment Analysis"
] |
[
67,
78
] |
http://arxiv.org/abs/2204.00796v1
|
A Dual-Contrastive Framework for Low-Resource Cross-Lingual Named Entity Recognition
|
Cross-lingual Named Entity Recognition (NER) has recently become a research hotspot because it can alleviate the data-hungry problem for low-resource languages. However, few researches have focused on the scenario where the source-language labeled data is also limited in some specific domains. A common approach for this scenario is to generate more training data through translation or generation-based data augmentation method. Unfortunately, we find that simply combining source-language data and the corresponding translation cannot fully exploit the translated data and the improvements obtained are somewhat limited. In this paper, we describe our novel dual-contrastive framework ConCNER for cross-lingual NER under the scenario of limited source-language labeled data. Specifically, based on the source-language samples and their translations, we design two contrastive objectives for cross-language NER at different grammatical levels, namely Translation Contrastive Learning (TCL) to close sentence representations between translated sentence pairs and Label Contrastive Learning (LCL) to close token representations within the same labels. Furthermore, we utilize knowledge distillation method where the NER model trained above is used as the teacher to train a student model on unlabeled target-language data to better fit the target language. We conduct extensive experiments on a wide variety of target languages, and the results demonstrate that ConCNER tends to outperform multiple baseline methods. For reproducibility, our code for this paper is available at https://github.com/GKLMIP/ConCNER.
|
[
"Multilinguality",
"Low-Resource NLP",
"Language Models",
"Machine Translation",
"Semantic Text Processing",
"Information Extraction & Text Mining",
"Representation Learning",
"Named Entity Recognition",
"Text Generation",
"Cross-Lingual Transfer",
"Responsible & Trustworthy NLP"
] |
[
0,
80,
52,
51,
72,
3,
12,
34,
47,
19,
4
] |
http://arxiv.org/abs/2109.03277v1
|
A Dual-Decoder Conformer for Multilingual Speech Recognition
|
Transformer-based models have recently become very popular for sequence-to-sequence applications such as machine translation and speech recognition. This work proposes a dual-decoder transformer model for low-resource multilingual speech recognition for Indian languages. Our proposed model consists of a Conformer [1] encoder, two parallel transformer decoders, and a language classifier. We use a phoneme decoder (PHN-DEC) for the phoneme recognition task and a grapheme decoder (GRP-DEC) to predict grapheme sequence along with language information. We consider phoneme recognition and language identification as auxiliary tasks in the multi-task learning framework. We jointly optimize the network for phoneme recognition, grapheme recognition, and language identification tasks with Joint CTC-Attention [2] training. Our experiments show that we can obtain a significant reduction in WER over the baseline approaches. We also show that our dual-decoder approach obtains significant improvement over the single decoder approach.
|
[
"Language Models",
"Semantic Text Processing",
"Information Retrieval",
"Information Extraction & Text Mining",
"Speech & Audio in NLP",
"Multimodality",
"Text Generation",
"Speech Recognition",
"Text Classification",
"Multilinguality"
] |
[
52,
72,
24,
3,
70,
74,
47,
10,
36,
0
] |
SCOPUS_ID:85135036721
|
A Dual-Expert Framework for Event Argument Extraction
|
Event argument extraction (EAE) is an important information extraction task, which aims to identify the arguments of an event described in a given text and classify the roles played by them. A key characteristic in realistic EAE data is that the instance numbers of different roles follow an obvious long-tail distribution. However, the training and evaluation paradigms of existing EAE models either prone to neglect the performance on "tail roles'', or change the role instance distribution for model training to an unrealistic uniform distribution. Though some generic methods can alleviate the class imbalance in long-tail datasets, they usually sacrifice the performance of "head classes'' as a trade-off. To address the above issues, we propose to train our model on realistic long-tail EAE datasets, and evaluate the average performance over all roles. Inspired by the Mixture of Experts (MOE), we propose a Routing-Balanced Dual Expert Framework (RBDEF), which divides all roles into "head"and "tail"two scopes and assigns the classifications of head and tail roles to two separate experts. In inference, each encoded instance will be allocated to one of the two experts by a routing mechanism. To reduce routing errors caused by the imbalance of role instances, we design a Balanced Routing Mechanism (BRM), which transfers several head roles to the tail expert to balance the load of routing, and employs a tri-filter routing strategy to reduce the misallocation of the tail expert's instances. To enable an effective learning of tail roles with scarce instances, we devise Target-Specialized Meta Learning (TSML) to train the tail expert. Different from other meta learning algorithms that only search a generic parameter initialization equally applying to infinite tasks, TSML can adaptively adjust its search path to obtain a specialized initialization for the tail expert, thereby expanding the benefits to the learning of tail roles. In experiments, RBDEF significantly outperforms the state-of-the-art EAE models and advanced methods for long-tail data.
|
[
"Language Models",
"Semantic Text Processing",
"Event Extraction",
"Argument Mining",
"Reasoning",
"Information Extraction & Text Mining"
] |
[
52,
72,
31,
60,
8,
3
] |
SCOPUS_ID:85102633576
|
A Dual-Index Based Representation for Processing XPath Queries on Very Large XML Documents
|
Although XML processing has been intensively studied in recent years, designing efficient implementations for evaluating XPath queries on XML documents remains a challenge in case XML documents are very large. In this study, we implemented a tree-shaped data structure called partial tree that is intrinsically suitable for large XML document processing with multiple computers. Our implementation uses two index sets to accelerate the evaluation of structural relationships among nodes, making it highly efficient for processing very large XML documents regarding three important classes of XPath queries: backward, order-aware and predicate-containing queries. Experiment results show that our implementation outperforms a start-of-the-art XML database BaseX in both absolute loading time and execution time for the target queries. The absolute execution time over 358 GB of XML data averagely is only seconds by using 32 EC2 instances.
|
[
"Semantic Text Processing",
"Green & Sustainable NLP",
"Representation Learning",
"Indexing",
"Information Retrieval",
"Responsible & Trustworthy NLP"
] |
[
72,
68,
12,
69,
24,
4
] |
SCOPUS_ID:85132723477
|
A Dual-Pointer guided transition system for end-to-end structured sentiment analysis with global graph reasoning
|
Structured sentiment analysis is a newly proposed task, which aims to summarize the overall sentiment and opinion status on given texts, i.e., the opinion expression, the sentiment polarity of the opinion, the holder of the opinion, and the target the opinion towards. In this work, we investigate a transition-based model for end-to-end structured sentiment analysis task. We design a transition architecture which supports the recognition of all the possible opinion quadruples in one shot. Based on the transition backbone, we then propose a Dual-Pointer module for more accurate term boundary detection. Besides, we further introduce a global graph reasoning mechanism, which helps to learn the global-level interactions between the overlapped quadruples. The high-order features are navigated into the transition system to enhance the final predictions. Extensive experimental results on five benchmarks demonstrate both the prominent efficacy and efficiency of our system. Our model outperforms all baselines in terms of all metrics, especially achieving a 10.5% point gain over the current best-performing system only detecting the holder-target-opinion triplets. Further analyses reveal that our framework is also effective in solving the overlapping structure and long-range dependency issues.
|
[
"Structured Data in NLP",
"Sentiment Analysis",
"Knowledge Graph Reasoning",
"Reasoning",
"Multimodality"
] |
[
50,
78,
54,
8,
74
] |
SCOPUS_ID:85084285674
|
A Dual-Purpose Refreshable Braille Display Based on Real Time Object Detection and Optical Character Recognition
|
This paper proposes a dual-purpose braille system for the visually impaired people. There are two main features of this system-object detection and optical character recognition. Real time object detection will help a visually impaired person to know about the things around him and optical character recognition will help him reading characters in both international (English) and local community (Bengali) language. In this paper, the detailed methodology of our proposed method is described. A pre-Trained convolutional neural network (AlexNet) is used for classifying the objects and an OCR engine (Tesseract) along with basic image processing is used for optical character recognition. A refreshable braille display is also designed to show the braille characters.
|
[
"Visual Data in NLP",
"Multimodality"
] |
[
20,
74
] |
http://arxiv.org/abs/2104.07221v1
|
A Dual-Questioning Attention Network for Emotion-Cause Pair Extraction with Context Awareness
|
Emotion-cause pair extraction (ECPE), an emerging task in sentiment analysis, aims at extracting pairs of emotions and their corresponding causes in documents. This is a more challenging problem than emotion cause extraction (ECE), since it requires no emotion signals which are demonstrated as an important role in the ECE task. Existing work follows a two-stage pipeline which identifies emotions and causes at the first step and pairs them at the second step. However, error propagation across steps and pair combining without contextual information limits the effectiveness. Therefore, we propose a Dual-Questioning Attention Network to alleviate these limitations. Specifically, we question candidate emotions and causes to the context independently through attention networks for a contextual and semantical answer. Also, we explore how weighted loss functions in controlling error propagation between steps. Empirical results show that our method performs better than baselines in terms of multiple evaluation metrics. The source code can be obtained at https://github.com/QixuanSun/DQAN.
|
[
"Information Extraction & Text Mining"
] |
[
3
] |
SCOPUS_ID:85141431321
|
A Dual-channel Text Classification Model based on an Interactive Attention Mechanism
|
Aiming at the problem that convolutional neural network(CNN) focuses on local features and lacks the ability of text context feature extraction, In this paper, we propose a dual-channel text classification model based on Interactive Attention Mechanism(IAM). The model uses skip-gram to embed words into dense low latitude vectors and obtains the text embedding matrix, which is input into the Gate Convolution Neural Network(GCNN) and Multi-Head Attention(MHA) at the same time, and then after Pointwise Convolution(PC), the features obtained from the feature extraction layer in the two channels are calculated by an IAM, and finally, the features are fused. Compared with CNN, LSTM, and other improved models, the classification effect of this hybrid model is improved.
|
[
"Information Retrieval",
"Text Classification",
"Information Extraction & Text Mining"
] |
[
24,
36,
3
] |
SCOPUS_ID:84959907412
|
A Dyadic Perspective on Speech Accommodation and Social Connection: Both Partners' Rejection Sensitivity Matters
|
Findings from confederate paradigms predict that mimicry is an adaptive route to social connection for rejection-sensitive individuals (Lakin, Chartrand, & Arkin, 2008). However, dyadic perspectives predict that whether mimicry leads to perceived connection depends on the rejection sensitivity (RS) of both partners in an interaction. We investigated these predictions in 50 college women who completed a dyadic cooperative task in which members were matched or mismatched in being dispositionally high or low in RS. We used a psycholinguistics paradigm to assess, through independent listeners' judgments (N=162), how much interacting individuals accommodate phonetic aspects of their speech toward each other. Results confirmed predictions from confederate paradigms in matched RS dyads. However, mismatched dyads showed an asymmetry in levels of accommodation and perceived connection: Those high in RS accommodated more than their low-RS partner but emerged feeling less connected. Mediational analyses indicated that low-RS individuals' nonaccommodation in mismatched dyads helped explain their high-RS partners' relatively low perceived connection to them. Establishing whether mimicry is an adaptive route to social connection requires analyzing mimicry as a dyadic process influenced by the needs of each dyad member.
|
[
"Linguistics & Cognitive NLP",
"Speech & Audio in NLP",
"Psycholinguistics",
"Multimodality"
] |
[
48,
70,
77,
74
] |
http://arxiv.org/abs/cmp-lg/9508007v1
|
A Dynamic Approach to Rhythm in Language: Toward a Temporal Phonology
|
It is proposed that the theory of dynamical systems offers appropriate tools to model many phonological aspects of both speech production and perception. A dynamic account of speech rhythm is shown to be useful for description of both Japanese mora timing and English timing in a phrase repetition task. This orientation contrasts fundamentally with the more familiar symbolic approach to phonology, in which time is modeled only with sequentially arrayed symbols. It is proposed that an adaptive oscillator offers a useful model for perceptual entrainment (or `locking in') to the temporal patterns of speech production. This helps to explain why speech is often perceived to be more regular than experimental measurements seem to justify. Because dynamic models deal with real time, they also help us understand how languages can differ in their temporal detail---contributing to foreign accents, for example. The fact that languages differ greatly in their temporal detail suggests that these effects are not mere motor universals, but that dynamical models are intrinsic components of the phonological characterization of language.
|
[
"Syntactic Text Processing",
"Phonology",
"Speech & Audio in NLP",
"Multimodality"
] |
[
15,
6,
70,
74
] |
SCOPUS_ID:85082583169
|
A Dynamic Bayesian Network Approach for Analysing Topic-Sentiment Evolution
|
Sentiment analysis is one of the key tasks of natural language understanding. Sentiment Evolution models the dynamics of sentiment orientation over time. It can help people have a more profound and deep understanding of opinion and sentiment implied in user generated content. Existing work mainly focuses on sentiment classification, while the analysis of how the sentiment orientation of a topic has been influenced by other topics or the dynamic interaction of topics from the aspect of sentiment has been ignored. In this paper, we propose to construct a Gaussian Process Dynamic Bayesian Network to model the dynamics and interactions of the sentiment of topics on social media such as Twitter. We use Dynamic Bayesian Networks to model time series of the sentiment of related topics and learn relationships between them. The network model itself applies Gaussian Process Regression to model the sentiment at a given time point based on related topics at previous time. We conducted experiments on a real world dataset that was crawled from Twitter with 9.72 million tweets. The experiment demonstrates a case study of analysing the sentiment dynamics of topics related to the event Brexit.
|
[
"Sentiment Analysis"
] |
[
78
] |
SCOPUS_ID:85034664290
|
A Dynamic Conditional Random Field Based Framework for Sentence-Level Sentiment Analysis of Chinese Microblog
|
With the increasing popularity of social media, the Sentiment Analysis (SA) of the Microblog has raised as a new research topic. In this paper, we present WDCRF: a Word2vec and Dynamic Conditional Random Field (DCRF) based framework for Sentiment Analysis of Chinese Microblog. Our contributions include: firstly, to address drawbacks of Microblog message such as the length and Lexicon limitations, Word2vec technology is leveraged to enrich Microblog message so that each word individual is extended by its Top-k similar words. Secondly, DCRF model is utilized to combine and conduct the Subjectivity Classification and Polarity Classification simultaneously, while in existing works they are designed as independent and the relationship between two types of classifications is ignored. Moreover, the DCRF model considers not only the classification-level relationship but also the relationship between neighboring sentences. Finally, the experiments on real dataset collected from Sina and Tencent Weibo demonstrate that our WDCRF (Word2vec + DCRF) achieves much better than the state-of-the-art.
|
[
"Information Extraction & Text Mining",
"Information Retrieval",
"Text Classification",
"Sentiment Analysis"
] |
[
3,
24,
36,
78
] |
SCOPUS_ID:85115269071
|
A Dynamic Convolutional Neural Network Approach for Legal Text Classification
|
The Amount of legal information that is being produced on a daily basis in courts is increasing enormously. The processing of such data has been receiving considerate attention thanks to their availability in an electronic form and the progress made in Artificial Intelligence application. Indeed, deep learning has shown promising results when used in the field of natural language processing (NLP). Neural Networks such as convolutional neural networks and recurrent neural network have been used for different NLP tasks like information retrieval, sentiment analysis and document classification. In this work, we propose a Neural Network based model with a dynamic input length for French legal text classification. The proposed approach, tested over real legal cases, outperforms baseline methods.
|
[
"Information Retrieval",
"Text Classification",
"Information Extraction & Text Mining"
] |
[
24,
36,
3
] |
SCOPUS_ID:85092187673
|
A Dynamic Emergency Decision-Making Method Based on Group Decision Making with Uncertainty Information
|
In emergency decision making (EDM), it is necessary to generate an effective alternative quickly. Case-based reasoning (CBR) has been applied to EDM; however, choosing the most suitable case from a set of similar cases after case retrieval remains challenging. This study proposes a dynamic method based on case retrieval and group decision making (GDM), called dynamic case-based reasoning group decision making (CBRGDM), for emergency alternative generation. In the proposed method, first, similar historical cases are identified through case similarity measurement. Then, evaluation information provided by group decision makers for similar cases is aggregated based on regret theory, and comprehensive perceived utilities for the similar cases are obtained. Finally, the most suitable historical case is obtained from the case similarities and the comprehensive perceived utilities for similar historical cases. The method is then applied to an example of a gas explosion in a coal company in China. The results show that the proposed method is feasible and effective in EDM. The advantages of the proposed method are verified based on comparisons with existing methods. In particular, dynamic CBRGDM can adjust the emergency alternative according to changing emergencies. The results of application of dynamic CBRGDM to a gas explosion and comparison with existing methods verify its feasibility and practicability.
|
[
"Reasoning",
"Information Retrieval"
] |
[
8,
24
] |
http://arxiv.org/abs/1905.05550v2
|
A Dynamic Evolutionary Framework for Timeline Generation based on Distributed Representations
|
Given the collection of timestamped web documents related to the evolving topic, timeline summarization (TS) highlights its most important events in the form of relevant summaries to represent the development of a topic over time. Most of the previous work focuses on fully-observable ranking models and depends on hand-designed features or complex mechanisms that may not generalize well. We present a novel dynamic framework for evolutionary timeline generation leveraging distributed representations, which dynamically finds the most likely sequence of evolutionary summaries in the timeline, called the Viterbi timeline, and reduces the impact of events that irrelevant or repeated to the topic. The assumptions of the coherence and the global view run through our model. We explore adjacent relevance to constrain timeline coherence and make sure the events evolve on the same topic with a global view. Experimental results demonstrate that our framework is feasible to extract summaries for timeline generation, outperforms various competitive baselines, and achieves the state-of-the-art performance as an unsupervised approach.
|
[
"Semantic Text Processing",
"Representation Learning"
] |
[
72,
12
] |
http://arxiv.org/abs/2108.01377v1
|
A Dynamic Head Importance Computation Mechanism for Neural Machine Translation
|
Multiple parallel attention mechanisms that use multiple attention heads facilitate greater performance of the Transformer model for various applications e.g., Neural Machine Translation (NMT), text classification. In multi-head attention mechanism, different heads attend to different parts of the input. However, the limitation is that multiple heads might attend to the same part of the input, resulting in multiple heads being redundant. Thus, the model resources are under-utilized. One approach to avoid this is to prune least important heads based on certain importance score. In this work, we focus on designing a Dynamic Head Importance Computation Mechanism (DHICM) to dynamically calculate the importance of a head with respect to the input. Our insight is to design an additional attention layer together with multi-head attention, and utilize the outputs of the multi-head attention along with the input, to compute the importance for each head. Additionally, we add an extra loss function to prevent the model from assigning same score to all heads, to identify more important heads and improvise performance. We analyzed performance of DHICM for NMT with different languages. Experiments on different datasets show that DHICM outperforms traditional Transformer-based approach by large margin, especially, when less training data is available.
|
[
"Language Models",
"Machine Translation",
"Semantic Text Processing",
"Text Generation",
"Multilinguality"
] |
[
52,
51,
72,
47,
0
] |
SCOPUS_ID:85116488127
|
A Dynamic Multi-criteria Multi-engine Approach for Text Simplification
|
In this work we present a multi-criteria multi-engine approach for text simplification. The main goal is to demonstrate a way to take advantage of a pool of systems, since in the literature several systems have been proposed for the task, and the results have been improving considerably. Note though, that such systems can behave differently, better or worse than the other ones, according to the input. For this reason, in this work we investigate the benefits of exploiting multiple systems at once, in a single-engine, in order to select the most appropriate simplification output from a pool of candidate outputs. In such an engine, a multi-critera decision making approach selects the final output considering simplicity and similarity scores, by comparing the candidates with the input. Results on both the Turk and WikiSmall corpora indicate that the proposed framework is able to balance the trade-off between bilingual evaluation understudy (BLEU), system output against references and against the input sentence (SARI), and Flesch reading ease scores for existing state-of-the art models.
|
[
"Paraphrasing",
"Text Generation"
] |
[
32,
47
] |
SCOPUS_ID:85097441331
|
A Dynamic Network Approach to the Study of Syntax
|
Usage-based linguists and psychologists have produced a large body of empirical results suggesting that linguistic structure is derived from language use. However, while researchers agree that these results characterize grammar as an emergent phenomenon, there is no consensus among usage-based scholars as to how the various results can be explained and integrated into an explicit theory or model. Building on network theory, the current paper outlines a structured network approach to the study of grammar in which the core concepts of syntax are analyzed by a set of relations that specify associations between different aspects of a speaker’s linguistic knowledge. These associations are shaped by domain-general processes that can give rise to new structures and meanings in language acquisition and language change. Combining research from linguistics and psychology, the paper proposes specific network analyses for the following phenomena: argument structure, word classes, constituent structure, constructions and construction families, and grammatical categories such as voice, case and number. The article builds on data and analyses presented in Diessel (2019; The Grammar Network. How Linguistic Structure is Shaped by Language Use) but approaches the topic from a different perspective.
|
[
"Linguistics & Cognitive NLP",
"Syntactic Text Processing",
"Linguistic Theories"
] |
[
48,
15,
57
] |
SCOPUS_ID:85043724057
|
A Dynamic Neural Field Model of Speech Cue Compensation
|
Categorical speech content can often be perceived directly from continuous auditory cues in the speech stream, but human-level performance on speech recognition tasks requires compensation for contextual variables like speaker identity. Regression modeling by McMurray and Jongman (2011) has suggested that for many fricative phonemes, a compensation scheme can substantially increase categorization accuracy beyond even the information from 24 un-compensated raw speech cues. Here, we simulate the same dataset instead using a neurally rather than abstractly implemented model: a hybrid dynamic neural field model and connectionist network. Our model achieved slightly lower accuracy than McMurray and Jongman's but similar accuracy patterns across most fricatives. Results also compared similarly to more recent models that were also less neurally instantiated but somewhat closer fitting to humans in accuracy. An even less abstracted model is an immediate future goal, as is expanding the present model to additional sensory modalities and constancy/compensation effects.
|
[
"Text Generation",
"Speech Recognition",
"Speech & Audio in NLP",
"Multimodality"
] |
[
47,
10,
70,
74
] |
http://arxiv.org/abs/1805.05202v2
|
A Dynamic Oracle for Linear-Time 2-Planar Dependency Parsing
|
We propose an efficient dynamic oracle for training the 2-Planar transition-based parser, a linear-time parser with over 99% coverage on non-projective syntactic corpora. This novel approach outperforms the static training strategy in the vast majority of languages tested and scored better on most datasets than the arc-hybrid parser enhanced with the SWAP transition, which can handle unrestricted non-projectivity.
|
[
"Syntactic Parsing",
"Syntactic Text Processing"
] |
[
28,
15
] |
SCOPUS_ID:85083327894
|
A Dynamic Parameter Enhanced Network for distant supervised relation extraction
|
Distant Supervised Relation Extraction (DSRE) is usually formulated as a problem about classifying a bag of sentences that contains two query entities into the predefined relation classes. Most existing methods consider those relation classes as distinct semantic categories while ignoring their potential connections to query entities. In this paper, we propose to leverage this connection to improve the relation extraction accuracy. Our key ideas are twofold: (1) For sentences belonging to the same relation class, the keywords to express the relation can vary according to the input query entities, i.e., style shift. To account for this style shift, the model can adjust its parameters in accordance with entity types. (2) Some relation classes are semantically similar, and the entity types appear in one relation may also appear in others. Therefore, it can be trained across different relation classes and further enhance those classes with few samples, i.e., long-tail relations. To unify these two arguments, we developed a novel Dynamic Parameter Enhanced Network (DPEN) for Relation Extraction, which introduces a parameter generator that can dynamically generates the network parameters according to the input query entity types and relation classes. By using this mechanism, the network can simultaneously handle the style shift problem and enhance the prediction accuracy for long-tail relations. Through extensive experiments, our method which is built on the top of the non-BERT-based or BERT-based models, can achieve superior performance over the state-of-the-art methods.
|
[
"Language Models",
"Relation Extraction",
"Semantic Text Processing",
"Information Extraction & Text Mining"
] |
[
52,
75,
72,
3
] |
SCOPUS_ID:85075163575
|
A Dynamic Perspective on Publics and Counterpublics: The Role of the Blogosphere in Pushing the Issue of Climate Change During the 2016 US Presidential Campaign
|
Climate change was hardly debated during the 2016 US presidential campaign. Against this background and building upon Fraser's concept of counterpublics (1990), this paper examines whether climate change advocates used the English-speaking blogosphere to push their positions forward. This study uses blog data starting from the Republican nomination of Donald Trump (20 July 2016) to Election Day (8 November 2016) and applies a computerized classification algorithm and topic-modeling techniques to explore, first, the salience of skeptic and advocate positions toward climate change in the English-speaking blogosphere and, second, with which topics these positions are most connected. The results show that the positions and topics of climate change advocates were more salient online than those of climate-change skeptics during the 2016 US presidential campaign. Thus, the study shows that the relation between different publics in societal discourses is not static but may change dynamically over time.
|
[
"Topic Modeling",
"Information Extraction & Text Mining"
] |
[
9,
3
] |
http://arxiv.org/abs/cs/0310041v1
|
A Dynamic Programming Algorithm for the Segmentation of Greek Texts
|
In this paper we introduce a dynamic programming algorithm to perform linear text segmentation by global minimization of a segmentation cost function which consists of: (a) within-segment word similarity and (b) prior information about segment length. The evaluation of the segmentation accuracy of the algorithm on a text collection consisting of Greek texts showed that the algorithm achieves high segmentation accuracy and appears to be very innovating and promissing.
|
[
"Programming Languages in NLP",
"Multimodality"
] |
[
55,
74
] |
https://aclanthology.org//W08-1125/
|
A Dynamic Programming Approach to Document Length Constraints
|
[
"Programming Languages in NLP",
"Text Generation",
"Multimodality"
] |
[
55,
47,
74
] |
|
https://aclanthology.org//W19-5943/
|
A Dynamic Strategy Coach for Effective Negotiation
|
Negotiation is a complex activity involving strategic reasoning, persuasion, and psychology. An average person is often far from an expert in negotiation. Our goal is to assist humans to become better negotiators through a machine-in-the-loop approach that combines machine’s advantage at data-driven decision-making and human’s language generation ability. We consider a bargaining scenario where a seller and a buyer negotiate the price of an item for sale through a text-based dialogue. Our negotiation coach monitors messages between them and recommends strategies in real time to the seller to get a better deal (e.g., “reject the proposal and propose a price”, “talk about your personal experience with the product”). The best strategy largely depends on the context (e.g., the current price, the buyer’s attitude). Therefore, we first identify a set of negotiation strategies, then learn to predict the best strategy in a given dialogue context from a set of human-human bargaining dialogues. Evaluation on human-human dialogues shows that our coach increases the profits of the seller by almost 60%.
|
[
"Natural Language Interfaces",
"Dialogue Systems & Conversational Agents"
] |
[
11,
38
] |
SCOPUS_ID:85029211108
|
A Dynamic Topic Model and Matrix Factorization-Based Travel Recommendation Method Exploiting Ubiquitous Data
|
The vast volumes of community-contributed geotagged photos (CCGPs) available on the Web can be utilized to make travel location recommendations. The sparsity of user location interactions makes it difficult to learn travel preferences, because a user usually visits only a limited number of travel locations. Static topic models can be used to solve the sparsity problem by considering user travel topics. However, all travel histories of a user are regarded as one document drawn from a set of static topics, ignoring the evolving of topics and travel preferences. In this paper, we propose a dynamic topic model (DTM) and matrix factorization (MF)-based travel recommendation method. A DTM is used to obtain the temporally fine-grained topic distributions (i.e., implicit topic information) of users and locations. In addition, a large amount of explicit information is extracted from the metadata and visual contents of CCGPs, check-ins, and point of interest categories datasets. The information is used to obtain user-user and location-location similarity information, which is imposed as two regularization terms to constraint MF. The proposed method is evaluated on a publicly available Flickr dataset. Experimental results demonstrate that the proposed method can generate significantly superior recommendations compared to other state-of-the-art travel location recommendation studies.
|
[
"Topic Modeling",
"Information Extraction & Text Mining"
] |
[
9,
3
] |
SCOPUS_ID:85075828562
|
A Dynamic Word Representation Model Based on Deep Context
|
The currently used word embedding techniques use fixed vectors to represent words without the concept of context and dynamics. This paper proposes a deep neural network CoDyWor to model the context of words so that words in different contexts have different vector representations of words. First of all, each layer of the model captures contextual information for each word of the input statement from different angles, such as grammatical information and semantic information, et al. Afterwards, different weights are assigned to each layer of the model through a multi-layered attention mechanism. At last, the information of each layer is integrated to form a dynamic word with contextual information to represent the vector. By comparing different models on the public dataset, it is found that the model’s accuracy in the task of logical reasoning has increased by 2.0%, F1 value in the task of named entity recognition has increased by 0.47%, and F1 value in the task of reading comprehension has increased by 2.96%. The experimental results demonstrate that this technology of word representation enhances the effect of the existing word representation.
|
[
"Named Entity Recognition",
"Information Extraction & Text Mining",
"Semantic Text Processing",
"Representation Learning"
] |
[
34,
3,
72,
12
] |
http://arxiv.org/abs/2205.12176v1
|
A Dynamic, Interpreted CheckList for Meaning-oriented NLG Metric Evaluation -- through the Lens of Semantic Similarity Rating
|
Evaluating the quality of generated text is difficult, since traditional NLG evaluation metrics, focusing more on surface form than meaning, often fail to assign appropriate scores. This is especially problematic for AMR-to-text evaluation, given the abstract nature of AMR. Our work aims to support the development and improvement of NLG evaluation metrics that focus on meaning, by developing a dynamic CheckList for NLG metrics that is interpreted by being organized around meaning-relevant linguistic phenomena. Each test instance consists of a pair of sentences with their AMR graphs and a human-produced textual semantic similarity or relatedness score. Our CheckList facilitates comparative evaluation of metrics and reveals strengths and weaknesses of novel and traditional metrics. We demonstrate the usefulness of CheckList by designing a new metric GraCo that computes lexical cohesion graphs over AMR concepts. Our analysis suggests that GraCo presents an interesting NLG metric worth future investigation and that meaning-oriented NLG metrics can profit from graph-based metric components using AMR.
|
[
"Semantic Text Processing",
"Structured Data in NLP",
"Semantic Similarity",
"Representation Learning",
"Explainability & Interpretability in NLP",
"Text Generation",
"Responsible & Trustworthy NLP",
"Multimodality"
] |
[
72,
50,
53,
12,
81,
47,
4,
74
] |
SCOPUS_ID:85116111438
|
A Entity Attention-based model for Entity Relation Classification for Chinese Literature Text
|
Entity relation classification is one of the basic tasks in natural language processing. The performance of the existing relational classification in Chinese literature text is not ideal. To address the issues, we propose a entity attention-based model for entity relation classification for Chinese literature text. Our proposed model extracts key information from entity by using attention mechanism, and filters out redundant information. In addition, we integrate entity type information into the model to help the model classify relation more reasonably. Experimental results show that the proposed model outperforms the state-of-the-art methods on Chinese literature text.
|
[
"Information Retrieval",
"Text Classification",
"Information Extraction & Text Mining"
] |
[
24,
36,
3
] |
SCOPUS_ID:85123106752
|
A Entity Relation Extraction Model with Enhanced Position Attention in Food Domain
|
Entity-relationship extraction is a fine-grained task for constructing a knowledge graph of food public opinion in the field of food public opinion, and it is also an important research topic in the field of current information extraction. This paper aims at the multi-entity-to-relationship problem that often occurs in food public opinion, the entity-relationship types are extracted from the BERT (Bidirectional Encoder Representation from Transformers) network model; In the bidirectional long short-term memory network (BLSTM), the entity-relationship types extracted by BERT model are integrated, and the semantic role attention mechanism based on position awareness is introduced to construct a model BERT-BLSTM-based entity-relationship extraction model for food public opinion at the same time. In this paper, comparative experiments were conducted on the food sentiment data set. The experimental results show that the accuracy of the BERT-BLSTM-based food sentiment entity-relationship extraction model proposed in this paper is 8.7 ~ 13.94% higher than several commonly used deep neural network models on the food sentiment data set, which verifies the rationality and effectiveness of the model proposed in this paper.
|
[
"Language Models",
"Semantic Text Processing",
"Relation Extraction",
"Sentiment Analysis",
"Information Extraction & Text Mining"
] |
[
52,
72,
75,
78,
3
] |
SCOPUS_ID:85029740048
|
A Europe of multiple flows: Contested discursive integration in trans-European transport infrastructure policy-making
|
This paper presents an examination of the extent to which discursive integration is accompanying the European integration process, by focusing on the development of trans-European transport infrastructure networks. Because they facilitate movement across nation-state borders, these networks are central to European integration and have in fact constituted a key EU policy issue for more than two decades. Some authors have argued that their development has been driven by a hegemonic discourse that promotes the production of a ‘Europe of Flows’: a single, uniform space underpinned by a vision of ‘frictionless’ mobility through inter-city networks. However, the existence of such a discourse is questionable given the variety of rationales that may potentially influence the development of this type of infrastructure. Their claim is evaluated by means of an in-depth empirical study of the policy process surrounding a high-speed rail line of EU relevance in the Spanish region of the Basque Country. The analysis of the discursive constructions mobilized in this process indicates that the discourse on a ‘Europe of Flows’ is better conceptualized as one of the several storylines associated with different scales through which a wider hegemonic discourse is articulated. Whilst the heterogeneity of this discourse did not fundamentally contradict the development of a trans-European high-speed rail line, it did result in a policy compromise according to the influence the different coalitions were able to exert in the policy process. The analysis largely demonstrates the importance of considering the multi-scalar discursive landscape of policy-making in order to understand trans-European infrastructure development.
|
[
"Discourse & Pragmatics",
"Semantic Text Processing"
] |
[
71,
72
] |
SCOPUS_ID:85148694879
|
A Event Extraction Method of Document-Level Based on the Self-attention Mechanism
|
Event extraction is an important task in the field of natural language processing. However, most of the existing event extraction techniques focus on sentence-level extraction, which inevitably ignores the contextual features of sentences and the occurrence of multiple event trigger words in the same sentence. Therefore, this paper mainly uses the multi-head self-attention mechanism to integrate text features from multiple dimensions and levels to achieve the task of event detection at the level of text. First, convolutional neural network combined with dynamic multi-pool strategy is used to extract sentence level features. Secondly, the discourse feature representation of full-text information fusion is obtained by multi-head self-attention mechanism model. Finally, using the classifier function to classify, and then detect the trigger word and category of the event. Experimental results show that the proposed method achieves good results in document-level event extraction.
|
[
"Event Extraction",
"Language Models",
"Semantic Text Processing",
"Information Extraction & Text Mining"
] |
[
31,
52,
72,
3
] |
SCOPUS_ID:85149245234
|
A FAISS-based Search for Story Generation
|
Stories have the power to change human perspectives and have applications in game development and film making. An intelligent system can generate appropriate stories for a set of keywords. We aim to build a system capable of getting stories by providing keywords as input. The stories must have a relation with the input keyword. We experimented with the ROCStory dataset. The preprocessed data are encoded using a sentence transformer, called msmarco-distilbert-base-prod-v3. We relied on a search approach based on Facebook AI Similarity Search (FAISS) to generate appropriate stories. The output story has been converted to audio via pyttsx3. The performance of the proposed model is compared with that of the sentence transformer paraphrase-MiniLM-L6-v2 approach. We made a subjective evaluation. The results show that the proposed approach outperforms the baseline method.
|
[
"Language Models",
"Semantic Text Processing",
"Information Retrieval",
"Text Generation"
] |
[
52,
72,
24,
47
] |
SCOPUS_ID:85135376883
|
A FAST MULTILINGUAL PROBABILISTIC TAGGER
|
This paper presents and compares two versions of a novel automatic tagging system which is both language and tagset independent and has close to real-time response in personal computers. The system's prediction model is based on the HMM chain theory and tags each word of a text, which includes also unknown words, using the Viterbi algorithm. The first version carries out floating-point arithmetic operations while the second version these operations have been transformed to fixed-point ones. Thus a significant time response reduction is achieved with negligible influence (<0.01%) on the prediction accuracy. The tagging system was tested on newspaper texts of 7 European languages using various sets of grammatical categories and texts with and without unknown words. The results proved to be satisfactory.
|
[
"Tagging",
"Syntactic Text Processing",
"Multilinguality"
] |
[
63,
15,
0
] |
SCOPUS_ID:85050754079
|
A FKSVM model based on Fisher criterion for text classification
|
Text classification is the process of automatically assigning a given document to its previous category. It is widely used in artificial intelligence and natural language processing. In this paper, we propose a new classification model named FKSVM in order to improve the accuracy of text classification. According to the model, firstly, we use the TF-IDF algorithm to calculate the weight of features, and then the feature vectors are sorted according to the value of T calculated by Anova and T-test. Secondly, the number of features that can produce the optimal classification performance is selected according to the Fisher criterion function. Finally, the KSVM classifier is used to classify which is the combination of SVM and KNN. The algorithm calculates the distance between the sample and the optimal hyperplane of SVM. If the distance is larger than the pre given threshold, the sample will be classified by SVM. Otherwise, the KNN classifier will be used. The experiment results indicate that the average accuracy of FKSVM model is up to 88.83%, while that of KNN is 83.20%, SVM is 84.73% and KSVM is 85.56%.
|
[
"Information Retrieval",
"Text Classification",
"Information Extraction & Text Mining"
] |
[
24,
36,
3
] |
http://arxiv.org/abs/1611.00801v1
|
A FOFE-based Local Detection Approach for Named Entity Recognition and Mention Detection
|
In this paper, we study a novel approach for named entity recognition (NER) and mention detection in natural language processing. Instead of treating NER as a sequence labelling problem, we propose a new local detection approach, which rely on the recent fixed-size ordinally forgetting encoding (FOFE) method to fully encode each sentence fragment and its left/right contexts into a fixed-size representation. Afterwards, a simple feedforward neural network is used to reject or predict entity label for each individual fragment. The proposed method has been evaluated in several popular NER and mention detection tasks, including the CoNLL 2003 NER task and TAC-KBP2015 and TAC-KBP2016 Tri-lingual Entity Discovery and Linking (EDL) tasks. Our methods have yielded pretty strong performance in all of these examined tasks. This local detection approach has shown many advantages over the traditional sequence labelling methods.
|
[
"Named Entity Recognition",
"Information Extraction & Text Mining"
] |
[
34,
3
] |
http://arxiv.org/abs/0905.0740v1
|
A FORTRAN coded regular expression Compiler for IBM 1130 Computing System
|
REC (Regular Expression Compiler) is a concise programming language which allows students to write programs without knowledge of the complicated syntax of languages like FORTRAN and ALGOL. The language is recursive and contains only four elements for control. This paper describes an interpreter of REC written in FORTRAN.
|
[
"Programming Languages in NLP",
"Multimodality"
] |
[
55,
74
] |
SCOPUS_ID:85142716877
|
A Face Recognition and Sentiment Analysis Activity System using Machine Learning Algorithm
|
As well known, there has always been a strong connection between the attendance of a student being linked to their performance which indeed refers totheir success ultimately, and seminars play an important rolewhen it comes to assisting the students to meet industries' expectations. And the traditional way of collecting the attendance before seminars is time-consuming, thus making less time for the presenter, and this can be solved using face recognitiontechnology. Whenever there is a seminar to be held, all eligible members should register themselves in the proposed activity system with all their credentials alongwith their facial image. As the seminar starts, the admin captures the images of the members who are attending the seminar, and these images are used for attendance by using face recognition technology. Here, the Histogram of Oriented Gradients is used for face detection and the K- Nearest Neighbors algorithm is used for face recognition which has shown the highest accuracy. The accuracy of the algorithm in this project has shown to be 97%. The proposed system along with providing the list of all attended members also provides the attended members with a feedback slot where the members can provide feedback in the form of text. Using Natural Language Processing, S entiment Analysis when performed onthe feedback gives a clear picture of the opinions of the members, represented in the form of a histogram showing the number of members who opted for positive or negative or neutral feedback.
|
[
"Visual Data in NLP",
"Multimodality",
"Sentiment Analysis"
] |
[
20,
74,
78
] |
SCOPUS_ID:85114636489
|
A Facial Landmark Detection Method Based on Deep Knowledge Transfer
|
Facial landmark detection is a crucial preprocessing step in many applications that process facial images. Deep-learning-based methods have become mainstream and achieved outstanding performance in facial landmark detection. However, accurate models typically have a large number of parameters, which results in high computational complexity and execution time. A simple but effective facial landmark detection model that achieves a balance between accuracy and speed is crucial. To achieve this, a lightweight, efficient, and effective model is proposed called the efficient face alignment network (EfficientFAN) in this article. EfficientFAN adopts the encoder-decoder structure, with a simple backbone EfficientNet-B0 as the encoder and three upsampling layers and convolutional layers as the decoder. Moreover, deep dark knowledge is extracted through feature-aligned distillation and patch similarity distillation on the teacher network, which contains pixel distribution information in the feature space and multiscale structural information in the affinity space of feature maps. The accuracy of EfficientFAN is further improved after it absorbs dark knowledge. Extensive experimental results on public datasets, including 300 Faces in the Wild (300W), Wider Facial Landmarks in the Wild (WFLW), and Caltech Occluded Faces in the Wild (COFW), demonstrate the superiority of EfficientFAN over state-of-the-art methods.
|
[
"Language Models",
"Responsible & Trustworthy NLP",
"Semantic Text Processing",
"Green & Sustainable NLP"
] |
[
52,
4,
72,
68
] |
https://aclanthology.org//2021.fever-1.13/
|
A Fact Checking and Verification System for FEVEROUS Using a Zero-Shot Learning Approach
|
In this paper, we propose a novel fact checking and verification system to check claims against Wikipedia content. Our system retrieves relevant Wikipedia pages using Anserini, uses BERT-large-cased question answering model to select correct evidence, and verifies claims using XLNET natural language inference model by comparing it with the evidence. Table cell evidence is obtained through looking for entity-matching cell values and TAPAS table question answering model. The pipeline utilizes zero-shot capabilities of existing models and all the models used in the pipeline requires no additional training. Our system got a FEVEROUS score of 0.06 and a label accuracy of 0.39 in FEVEROUS challenge.
|
[
"Language Models",
"Low-Resource NLP",
"Semantic Text Processing",
"Structured Data in NLP",
"Question Answering",
"Multimodality",
"Natural Language Interfaces",
"Ethical NLP",
"Reasoning",
"Fact & Claim Verification",
"Responsible & Trustworthy NLP"
] |
[
52,
80,
72,
50,
27,
74,
11,
17,
8,
46,
4
] |
http://arxiv.org/abs/1803.00712v3
|
A Factoid Question Answering System for Vietnamese
|
In this paper, we describe the development of an end-to-end factoid question answering system for the Vietnamese language. This system combines both statistical models and ontology-based methods in a chain of processing modules to provide high-quality mappings from natural language text to entities. We present the challenges in the development of such an intelligent user interface for an isolating language like Vietnamese and show that techniques developed for inflectional languages cannot be applied "as is". Our question answering system can answer a wide range of general knowledge questions with promising accuracy on a test set.
|
[
"Natural Language Interfaces",
"Question Answering"
] |
[
11,
27
] |
SCOPUS_ID:85117443057
|
A Factoid based Question Answering System based on Dependency Analysis and Wikidata
|
Over the last years, the use and the need for automated question answering systems have become more important than ever. The main reasons for this relate to the constant increase of the information that is available in textual form as well as the need to facilitate users in getting information they seek in a precise, fast, and easy way. In this work, we deal with the problem of the open domain factoid-based answering of questions, where the answer to a question is in the form of a word or a small phrase. We present an efficient methodology for analyzing questions and specifying proper answers with the use of knowledge bases. Initially, given a specific question that a user sets, a high-quality linguistic analysis of the question is performed. The dependencies of the question are specified, and the main triplets of the question are created. The system creates a SPARQL query to get the right answer for the question via the API from the Wikidata Query Service. The results are quite encouraging and highlight the quite good performance of our methodology.
|
[
"Natural Language Interfaces",
"Question Answering"
] |
[
11,
27
] |
http://arxiv.org/abs/1604.05878v1
|
A Factorization Machine Framework for Testing Bigram Embeddings in Knowledgebase Completion
|
Embedding-based Knowledge Base Completion models have so far mostly combined distributed representations of individual entities or relations to compute truth scores of missing links. Facts can however also be represented using pairwise embeddings, i.e. embeddings for pairs of entities and relations. In this paper we explore such bigram embeddings with a flexible Factorization Machine model and several ablations from it. We investigate the relevance of various bigram types on the fb15k237 dataset and find relative improvements compared to a compositional model.
|
[
"Semantic Text Processing",
"Representation Learning"
] |
[
72,
12
] |
http://arxiv.org/abs/1602.01576v1
|
A Factorized Recurrent Neural Network based architecture for medium to large vocabulary Language Modelling
|
Statistical language models are central to many applications that use semantics. Recurrent Neural Networks (RNN) are known to produce state of the art results for language modelling, outperforming their traditional n-gram counterparts in many cases. To generate a probability distribution across a vocabulary, these models require a softmax output layer that linearly increases in size with the size of the vocabulary. Large vocabularies need a commensurately large softmax layer and training them on typical laptops/PCs requires significant time and machine resources. In this paper we present a new technique for implementing RNN based large vocabulary language models that substantially speeds up computation while optimally using the limited memory resources. Our technique, while building on the notion of factorizing the output layer by having multiple output layers, improves on the earlier work by substantially optimizing on the individual output layer size and also eliminating the need for a multistep prediction process.
|
[
"Language Models",
"Semantic Text Processing"
] |
[
52,
72
] |
http://arxiv.org/abs/1911.01460v1
|
A Failure of Aspect Sentiment Classifiers and an Adaptive Re-weighting Solution
|
Aspect-based sentiment classification (ASC) is an important task in fine-grained sentiment analysis.~Deep supervised ASC approaches typically model this task as a pair-wise classification task that takes an aspect and a sentence containing the aspect and outputs the polarity of the aspect in that sentence. However, we discovered that many existing approaches fail to learn an effective ASC classifier but more like a sentence-level sentiment classifier because they have difficulty to handle sentences with different polarities for different aspects.~This paper first demonstrates this problem using several state-of-the-art ASC models. It then proposes a novel and general adaptive re-weighting (ARW) scheme to adjust the training to dramatically improve ASC for such complex sentences. Experimental results show that the proposed framework is effective \footnote{The dataset and code are available at \url{https://github.com/howardhsu/ASC_failure}.}.
|
[
"Text Classification",
"Aspect-based Sentiment Analysis",
"Sentiment Analysis",
"Information Retrieval",
"Information Extraction & Text Mining"
] |
[
36,
23,
78,
24,
3
] |
SCOPUS_ID:85114774687
|
A Fair and Comprehensive Comparison of Multimodal Tweet Sentiment Analysis Methods
|
Opinion and sentiment analysis is a vital task to characterize subjective information in social media posts. In this paper, we present a comprehensive experimental evaluation and comparison with six state-of-the-art methods, from which we have re-implemented one of them. In addition, we investigate different textual and visual feature embeddings that cover different aspects of the content, as well as the recently introduced multimodal CLIP embeddings. Experimental results are presented for two different publicly available benchmark datasets of tweets and corresponding images. In contrast to the evaluation methodology of previous work, we introduce a reproducible and fair evaluation scheme to make results comparable. Finally, we conduct an error analysis to outline the limitations of the methods and possibilities for the future work.
|
[
"Semantic Text Processing",
"Representation Learning",
"Ethical NLP",
"Sentiment Analysis",
"Responsible & Trustworthy NLP",
"Multimodality"
] |
[
72,
12,
17,
78,
4,
74
] |
SCOPUS_ID:85141788467
|
A Fake News Detection System based on Combination of Word Embedded Techniques and Hybrid Deep Learning Model
|
At present, most people prefer using different online sources for reading news. These sources can easily spread fake news for several malicious reasons. Detecting this unreliable news is an important task in the Natural Language Processing (NLP) field. Many governments and technology companies are engaged in this research field to prevent the manipulation of public opinion and spare people and society the huge damage that can result from the spreading of misleading information on online social media. In this paper, we present a new deep learning method to detect fake news based on a combination of different word embedding techniques and a hybrid Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BILSTM) model. We trained the classification model on the unbiased dataset WELFake. The best method was a combination of a pre-trained Word2Vec CBOW model and a Word2Vec Skip-Word model with a CNN on BILSTM layers, yielding an accuracy of up to 97%.
|
[
"Language Models",
"Semantic Text Processing",
"Ethical NLP",
"Reasoning",
"Fact & Claim Verification",
"Responsible & Trustworthy NLP"
] |
[
52,
72,
17,
8,
46,
4
] |
SCOPUS_ID:85130247821
|
A Fake News Detection and Credibility Ranking Platform for Portuguese Online News
|
The growth of social media has enabled the spread of tendentiously fake news content in a disorganized and fast manner. Despite the extensive research on fake news detection methods, algorithms and applications [1], most of the studies focused on Natural Language Processing (NLP) techniques for English content analysis. Assessing online news in Portuguese language and ranking them based on their credibility is still an open issue. Addressing this gap, the present work proposes a platform for European Portuguese online news credibility classification and ranking based on content and publication metadata. A dataset was built from the collection of online news from 36 sources, comprising content-related data (title, lead, body) and supporting metadata (source, category, year, author). A hierarchical architecture was envisioned and implemented for supervised binary classification probabilities extraction aggregated into multi-classes, resorting to three Machine Learning models: i) focused on textual analysis resorting to several NLP techniques; ii) delving into the news publication metadata; and iii) leveraging the balance between the two prior models’ probability predictions. This architecture minimizes data entropy, maximizing information retrieved, providing two credibility indexes and a fake news detector with 92% test accuracy and 95% deployed proof-of-concept accuracy. These models were integrated into a web service, enabling users to evaluate and rank online news resorting to the implemented architecture. This solution has several application areas, fostering society’s literacy and increased critical thinking, especially amongst younger generations.
|
[
"Information Extraction & Text Mining",
"Information Retrieval",
"Ethical NLP",
"Reasoning",
"Fact & Claim Verification",
"Text Classification",
"Responsible & Trustworthy NLP"
] |
[
3,
24,
17,
8,
46,
36,
4
] |
https://aclanthology.org//2021.iwpt-1.8/
|
A Falta de Pan, Buenas Son Tortas: The Efficacy of Predicted UPOS Tags for Low Resource UD Parsing
|
We evaluate the efficacy of predicted UPOS tags as input features for dependency parsers in lower resource settings to evaluate how treebank size affects the impact tagging accuracy has on parsing performance. We do this for real low resource universal dependency treebanks, artificially low resource data with varying treebank sizes, and for very small treebanks with varying amounts of augmented data. We find that predicted UPOS tags are somewhat helpful for low resource treebanks, especially when fewer fully-annotated trees are available. We also find that this positive impact diminishes as the amount of data increases.
|
[
"Low-Resource NLP",
"Syntactic Text Processing",
"Syntactic Parsing",
"Tagging",
"Responsible & Trustworthy NLP"
] |
[
80,
15,
28,
63,
4
] |
SCOPUS_ID:56749149816
|
A Farsi part-of-speech tagger based on Markov model
|
This paper describes applying a Part-Of-Speech (POS) tagging system on an unreported Farsi corpus by using a Markov model. Some aspects of Farsi morphology and some issues in developing a tagging system are offered. By simulation we evaluate this method on the corpus. To our knowledge, this is first time that a statistical POS tagger is applied on a Farsi corpus. Copyright 2008 ACM.
|
[
"Tagging",
"Syntactic Text Processing"
] |
[
63,
15
] |
SCOPUS_ID:85043594170
|
A Fast Algorithm for Posterior Inference with Latent Dirichlet Allocation
|
Latent Dirichlet Allocation (LDA) [1], among various forms of topic models, is an important probabilistic generative model for analyzing large collections of text corpora. The problem of posterior inference for individual texts is very important in streaming environments, but is often intractable in the worst case. To avoid directly solving this problem which is NP-hard, some proposed existing methods for posterior inference are approximate but do not have any guarantee on neither quality nor convergence rate. Based on the idea of Online Frank-Wolfe algorithm by Hazan [2] and improvement of Online Maximum a Posteriori Estimation algorithm (OPE) by Than [3, 4], we propose a new effective algorithm (so-called NewOPE) solving posterior inference in topic models by combining Bernoulli distribution, stochastic bounds, and approximation function. Our algorithm has more attractive properties than existing inference approaches, including theoretical guarantees on quality and fast convergence rate. It not only maintains the key advantages of OPE but often outperforms OPE and existing algorithms before. Our new algorithm has been employed to develop two effective methods for learning topic models from massive/streaming text collections. Experimental results show that our approach is more efficient and robust than the state-of-the-art methods.
|
[
"Topic Modeling",
"Information Extraction & Text Mining"
] |
[
9,
3
] |
SCOPUS_ID:85009874545
|
A Fast Asymmetric Extremum Content Defined Chunking Algorithm for Data Deduplication in Backup Storage Systems
|
Chunk-level deduplication plays an important role in backup storage systems. Existing Content-Defined Chunking (CDC) algorithms, while robust in finding suitable chunk boundaries, face the key challenges of (1) low chunking throughput that renders the chunking stage a serious deduplication performance bottleneck, (2) large chunk size variance that decreases deduplication efficiency, and (3) being unable to find proper chunk boundaries in low-entropy strings and thus failing to deduplicate these strings. To address these challenges, this paper proposes a new CDC algorithm called the Asymmetric Extremum (AE) algorithm. The main idea behind AE is based on the observation that the extreme value in an asymmetric local range is not likely to be replaced by a new extreme value in dealing with the boundaries-shifting problem. As a result, AE has higher chunking throughput, smaller chunk size variance than the existing CDC algorithms, and is able to find proper chunk boundaries in low-entropy strings. The experimental results based on real-world datasets show that AE improves the throughput performance of the state-of-the-art CDC algorithms by more than 2.3 × , which is fast enough to remove the chunking-throughput performance bottleneck of deduplication, and accelerates the system throughput by more than 50 percent, while achieving comparable deduplication efficiency.
|
[
"Responsible & Trustworthy NLP",
"Chunking",
"Syntactic Text Processing",
"Green & Sustainable NLP"
] |
[
4,
43,
15,
68
] |
http://arxiv.org/abs/2205.07646v1
|
A Fast Attention Network for Joint Intent Detection and Slot Filling on Edge Devices
|
Intent detection and slot filling are two main tasks in natural language understanding and play an essential role in task-oriented dialogue systems. The joint learning of both tasks can improve inference accuracy and is popular in recent works. However, most joint models ignore the inference latency and cannot meet the need to deploy dialogue systems at the edge. In this paper, we propose a Fast Attention Network (FAN) for joint intent detection and slot filling tasks, guaranteeing both accuracy and latency. Specifically, we introduce a clean and parameter-refined attention module to enhance the information exchange between intent and slot, improving semantic accuracy by more than 2%. FAN can be implemented on different encoders and delivers more accurate models at every speed level. Our experiments on the Jetson Nano platform show that FAN inferences fifteen utterances per second with a small accuracy drop, showing its effectiveness and efficiency on edge devices.
|
[
"Semantic Text Processing",
"Semantic Parsing",
"Intent Recognition",
"Natural Language Interfaces",
"Sentiment Analysis",
"Dialogue Systems & Conversational Agents"
] |
[
72,
40,
79,
11,
78,
38
] |
http://arxiv.org/abs/1802.10078v1
|
A Fast Deep Learning Model for Textual Relevance in Biomedical Information Retrieval
|
Publications in the life sciences are characterized by a large technical vocabulary, with many lexical and semantic variations for expressing the same concept. Towards addressing the problem of relevance in biomedical literature search, we introduce a deep learning model for the relevance of a document's text to a keyword style query. Limited by a relatively small amount of training data, the model uses pre-trained word embeddings. With these, the model first computes a variable-length Delta matrix between the query and document, representing a difference between the two texts, which is then passed through a deep convolution stage followed by a deep feed-forward network to compute a relevance score. This results in a fast model suitable for use in an online search engine. The model is robust and outperforms comparable state-of-the-art deep learning approaches.
|
[
"Information Retrieval"
] |
[
24
] |
SCOPUS_ID:85082167856
|
A Fast Estimation of Initial Rotor Position for Low-Speed Free-Running IPMSM
|
Fast and reliable initial rotor position detection is essential for restarting sensorless permanent magnet synchronous motors (PMSMs) in free-running condition. In this article, a fast initial rotor position estimation method for low-speed free-running motor is proposed, which utilizes a combined sinusoidal current and square-wave voltage injection method. The sinusoidal current is imposed into the estimated d-axis to magnify the magnetic saturation effect. The amplitudes of the d-axis current caused by injected square-wave voltage are then accumulated. The large difference of the two integrated signals for positive and negative d-axis currents can be reliably used to identify the rotor polarity. Meanwhile, in low-speed free-running stage, the change of saturation degrees introduced by the injected sinusoidal signal does not affect the position estimation accuracy. Moreover, even if the sinusoidal current signal is injected in the incorrect d-axis, the resultant torque is small and unexpected rotation of the rotor is prevented. Its influence on the free-running motor is negligible, due to the combined injection with continuously online updated estimated rotor position by high-frequency square-wave voltage injection during the polarity identification process. Finally, the effectiveness of the proposed method is investigated on a 1.5 kW interior PMSM test platform.
|
[
"Polarity Analysis",
"Sentiment Analysis"
] |
[
33,
78
] |
SCOPUS_ID:85132713369
|
A Fast Indoor Positioning Using a Knowledge-Distilled Convolutional Neural Network (KD-CNN)
|
Fingerprint-based indoor positioning systems (F-IPS) may provide inexpensive solutions to GPS-denied environments. Most F-IPSs adopt traditional machine learning for position prediction, resulting in low accuracy. Deep neural networks (DNN) were recently employed for F-IPSs to minimize positioning errors. Nevertheless, a DNN-IPS fails to guarantee high accuracy in dynamic environments as it is sensitive to changes in the input data. A convolutional neural network (CNN) is recommended to replace DNN due to its capability to learn the overall topology of fingerprinting images and capture highly abstract features. Due to the convolution process and image representation, CNN-IPS incurs prohibitive storage and computational requirement for implementation on resource-limited devices. This paper incorporates knowledge distillation (KD) into CNN-IPS to distil knowledge from large deep CNNs into small CNNs. The pre-trained teacher network uses the soft probability output where the score vector from the trained network is converted into a probability distribution, which is softened by the temperature hyperparameter, leading to a more simplified model. Based on the numerical results, KD-CNN-IPS manifests better localization performance where 79.84% of the positioning errors are within 2 meters while its testing time is only 79.68% of that of the teacher model. Compared to the CNN-IPS, KD-CNN-IPS with precisely the same architecture and size could achieve a performance improvement of 13.65% in terms of the average positioning error.
|
[
"Visual Data in NLP",
"Language Models",
"Semantic Text Processing",
"Green & Sustainable NLP",
"Responsible & Trustworthy NLP",
"Multimodality"
] |
[
20,
52,
72,
68,
4,
74
] |
SCOPUS_ID:85087029453
|
A Fast Mode of Tweets Polarity Detection
|
Polarity detection is an emerging area of research in text mining. Polarity detection is observing and identifying the sentiment inclination of text, whether it is positive or negative. In this paper, a fast mode of supervised learning for polarity detection on tweets is proposed, that is using datasets available in public. The feature selection strategy ensures reduced dimensionality. The low dimension data processing on Apache Spark supports scalability for large datasets. The experiment shows that the method is supporting high scalability and efficiency.
|
[
"Polarity Analysis",
"Sentiment Analysis"
] |
[
33,
78
] |
SCOPUS_ID:85041829184
|
A Fast Multi-level Plagiarism Detection Method Based on Document Embedding Representation
|
Nowadays, global networks facilitate access to vast amount of textual information and enhance the feasibility of plagiarism as a consequence. Given the amount of text material produced everyday, the need for an automated fast plagiarism detection system is more crucial than ever. Plagiarism detection is defined as identification of reused text materials. In this regard, different algorithms have been proposed to perform the task of plagiarism detection in text documents. Due to limitation in semantic representation and computational inefficiency of traditional algorithms for plagiarism detection, in this paper, we proposed an embedding based document representation to detect plagiarism in documents using a two-level decision making approach. The method is language-independent and works properly on various languages as well. In the proposed method, words are represented as multi-dimensional vectors, and simple aggregation methods are used to combine the word vectors in order to represent sentences. By comparing representations of source and suspicious sentences, sentence pairs with the highest similarity score are considered as the candidates of the plagiarism cases. The final decision whether or not the pairs are plagiarized is taken using another level of similarity calculation using Jaccard metric by comparing the word sets of two sentences. Our method has been used in PAN2016 Persian plagiarism detection contest and results in 85.8% recall, 95.9% precision and 90.6% plagdet which is a combination of the these two measures with the measure of how concretely we retrieve plagiarism cases, on the provided data sets in a short amount of time. This method achieved the second place regarding plagdet and the first rank based on runtime.
|
[
"Semantic Text Processing",
"Representation Learning"
] |
[
72,
12
] |
http://arxiv.org/abs/2204.09656v2
|
A Fast Post-Training Pruning Framework for Transformers
|
Pruning is an effective way to reduce the huge inference cost of Transformer models. However, prior work on pruning Transformers requires retraining the models. This can add high training cost and high complexity to model deployment, making it difficult to use in many practical situations. To address this, we propose a fast post-training pruning framework for Transformers that does not require any retraining. Given a resource constraint and a sample dataset, our framework automatically prunes the Transformer model using structured sparsity methods. To retain high accuracy without retraining, we introduce three novel techniques: (i) a lightweight mask search algorithm that finds which heads and filters to prune based on the Fisher information; (ii) mask rearrangement that complements the search algorithm; and (iii) mask tuning that reconstructs the output activations for each layer. We apply our method to BERT-base and DistilBERT, and we evaluate its effectiveness on GLUE and SQuAD benchmarks. Our framework achieves up to 2.0x reduction in FLOPs and 1.56x speedup in inference latency, while maintaining < 1% loss in accuracy. Importantly, our framework prunes Transformers in less than 3 minutes on a single GPU, which is over two orders of magnitude faster than existing pruning approaches that retrain the models.
|
[
"Language Models",
"Semantic Text Processing"
] |
[
52,
72
] |
SCOPUS_ID:85102010507
|
A Fast Search System for Remote Sensing Imagery Based on Bag of Visual Words and Latent Dirichlet Allocation
|
In this paper, we present our image search system for remote sensing imagery leveraging the capabilities of Elasticsearch, a well-known full-text search engine. We use bag of visual words and bag of visual topics model to represent the earth observation images in a text- equivalent format. The image files are stored in Elasticsearch in their text-equivalent format, making it feasible to apply some sophisticated text-based search functionalities of Elasticsearch on the data. This gives great flexibility to the search clauses allowing the user to match chosen fragments of images. We also implement visual topic search which enables search by higher level semantics. The proposed methods are simple and fast as they do not require any complex image feature computation.
|
[
"Visual Data in NLP",
"Information Retrieval",
"Multimodality"
] |
[
20,
24,
74
] |
SCOPUS_ID:85047193555
|
A Fast Uyghur Text Detector for Complex Background Images
|
Uyghur text localization in images with complex backgrounds is a challenging yet important task for many applications. Generally, Uyghur characters in images consist of strokes with uniform features, and they are distinct from backgrounds in color, intensity, and texture. Based on these differences, we propose a FASTroke keypoint extractor, which is fast and stroke-specific. Compared with the commonly used MSER detector, FASTroke produces less than twice the amount of components and recognizes at least 10% more characters. While the characters in a line usually have uniform features such as size, color, and stroke width, a component similarity based clustering is presented without component-level classification. It incurs no extra errors by incorporating a component-level classifier while the computing cost is drastically reduced. The experiments show that the proposed method can achieve the best performance on the UICBI-500 benchmark dataset.
|
[
"Visual Data in NLP",
"Multimodality"
] |
[
20,
74
] |
https://aclanthology.org//W06-1008/
|
A Fast and Accurate Method for Detecting English-Japanese Parallel Texts
|
[
"Multilinguality"
] |
[
0
] |
|
SCOPUS_ID:85044994675
|
A Fast and Accurate Rule-Base Generation Method for Mamdani Fuzzy Systems
|
The problem of learning fuzzy rule bases is analyzed from the perspective of finding a favorable balance between the accuracy of the system, the speed required to learn the rules, and, finally, the interpretability of the rule bases obtained. Therefore, we introduce a complete design procedure to learn and then optimize the system rule base, called the precise and fast fuzzy modeling approach. Under this paradigm, fuzzy rules are generated from numerical data using a parameterizable greedy-based learning method called selection-reduction, whose accuracy-speed efficiency is confirmed through empirical results and comparisons with reference methods. Qualitative justification for this method is provided based on the coaction between fuzzy logic and the intrinsic properties of greedy algorithms. To complete the precise and fast fuzzy modeling strategy, we finally present a rule-base optimization technique driven by a novel rule redundancy index, which takes into account the concepts of the distance between rules and the influence of a rule over the dataset. Experimental results show that the proposed index can be used to obtain compact rule bases, which remain very accurate, thus increasing system interpretability.
|
[
"Explainability & Interpretability in NLP",
"Responsible & Trustworthy NLP"
] |
[
81,
4
] |
http://arxiv.org/abs/1709.06307v2
|
A Fast and Accurate Vietnamese Word Segmenter
|
We propose a novel approach to Vietnamese word segmentation. Our approach is based on the Single Classification Ripple Down Rules methodology (Compton and Jansen, 1990), where rules are stored in an exception structure and new rules are only added to correct segmentation errors given by existing rules. Experimental results on the benchmark Vietnamese treebank show that our approach outperforms previous state-of-the-art approaches JVnSegmenter, vnTokenizer, DongDu and UETsegmenter in terms of both accuracy and performance speed. Our code is open-source and available at: https://github.com/datquocnguyen/RDRsegmenter.
|
[
"Text Segmentation",
"Syntactic Text Processing"
] |
[
21,
15
] |
SCOPUS_ID:85013046481
|
A Fast and Efficient Framework for Creating Parallel Corpus
|
Objectives: A framework involving Scansnap SV600 scanner and Google Optical character recognition (OCR) for creating parallel corpus which is a very essential component of Statistical Machine Translation (SMT). Methods and Analysis: Training a language model for a SMT system highly depends on the availability of a parallel corpus. An efficacious approach for collecting parallel sentences is the predominant step in an MT system. However, the creation of a parallel corpus requires extensive knowledge in both languages which is a time consuming process. Due to these limitations, making the documents digital becomes very difficult and which in turn affects the quality of machine translation systems. In this paper, we propose a faster and efficient way of generating English to Indian languages parallel corpus with less human involvement. With the help of a special type of scanner called Scansnap SV600 and Google OCR and a little linguistic knowledge, we can create a parallel corpus for any language pair, provided there should be paper documents with parallel sentences. Findings: It was possible to generate 40 parallel sentences in 1 hour time with this approach. Sophisticated morphological tools were used for changing the morphology of the text generated and thereby increase the size of the corpus. An additional benefit of this is to make ancient scriptures or other manuscripts in digital format which can then be referred by the coming generation to keep up the traditions of a nation or a society. Novelty: Time required for creating parallel corpus is reduced by incorporating Google OCR and book scanner.
|
[
"Visual Data in NLP",
"Machine Translation",
"Green & Sustainable NLP",
"Multimodality",
"Text Generation",
"Responsible & Trustworthy NLP",
"Multilinguality"
] |
[
20,
51,
68,
74,
47,
4,
0
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.