ClassiFin_dataset / README.md
librarian-bot's picture
Librarian Bot: Add language metadata for dataset
3f83ac2 verified
|
raw
history blame
3.44 kB
metadata
language:
  - en
dataset_info:
  features:
    - name: score
      dtype: int64
    - name: example_id
      dtype: string
    - name: text
      dtype: string
    - name: url
      dtype: string
    - name: timestamp
      dtype: string
    - name: source
      dtype: string
  splits:
    - name: train
      num_bytes: 398647381.09090906
      num_examples: 108000
    - name: test
      num_bytes: 7382358.909090909
      num_examples: 2000
  download_size: 250483279
  dataset_size: 406029740
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*

ClassiFin_dataset

Overview

ClassiFin_dataset is a dataset created for training and evaluating the ClassiFin model, designed to classify texts based on their relevance to finance, business, or economic topics. The dataset consists of 110,000 annotated documents sourced from the CulturaX dataset, with annotations provided by Qwen/Qwen2.5-72B-Instruct.

Dataset Structure

Each document in the dataset has been assigned a score between 0 and 5, reflecting its relevance to finance, business, or economics. The scoring is based on predefined criteria that assess the depth and complexity of financial and business content in the text. Additionally, each annotation includes a brief justification and a boolean flag indicating whether the text is finance-related.

Annotation Methodology

The dataset was annotated using the following prompt:

Below is a text from a dataset. Evaluate whether the text is related to finance, business, or economy topics using the following scoring system:

- Award 1 point if the Text contains basic financial/business terminology or concepts
- Add another point if the Text specifically asks about financial analysis, business strategies, or economic principles
- Award a third point if the Text demonstrates advanced understanding of financial/business concepts
- Grant a fourth point if the Text requires expert-level financial/business knowledge
- Award a fifth point if the Text is highly technical and specific to finance/business/economy

Text: {text}

Please provide:
1. A score from 0-5 based on the criteria above
2. A brief justification (max 50 words)
3. A boolean (true/false) indicating if this is finance/business related (score >= 1)

Respond with only a JSON object (do not add ```json around the json) with these exact keys:
{"score": <0-5 integer>,
"justification": "<50 char max explanation>",
"is_finance": <true if score >= 3, else false>}

Dataset Fields

Each sample in the dataset includes these two main fields:

  • text: The original document text from CulturaX.
  • score: An integer between 0-5 indicating financial relevance.

Besides there are several other mata information for each sample.

License

Please refer to both below licenses when using this dataset.

Citation

If you use this dataset in your research or projects, please cite it appropriately.

@misc{ClassiFin,
  title={ClassiFin: Finance Document Classifier},
  author={Liu, Jingshu and Qader, Raheel and Caillaut, Gaëtan and Nakhlem, Mariam and Barthelemy, Jean-Gabriel and Sadoune, Arezki and Foly, Sabine},
  url={https://huggingface.co/LinguaCustodia/ClassiFin},
  year={2025}
}