Duplicate from viewervoice-analytics/dev

.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

.streamlit/config.toml

@@ -0,0 +1,4 @@
+[theme]
+base="light"
+primaryColor="#2b67e0"
+backgroundColor="#eaf8ff"

README.md

@@ -0,0 +1,42 @@
+---
+emoji: 🤔
+sdk: streamlit
+duplicated_from: viewervoice-analytics/dev
+title: dev
+colorFrom: blue
+colorTo: green
+---
+# ViewerVoice
+
+<p align="center">
+<img width="620" alt="image" src="https://cdn-uploads.huggingface.co/production/uploads/649afd299cc39ae39ce9cc04/xWZdzLZtbtYFJFpYhD_sr.png">
+</p>
+
+## 🏁Introduction  
+Welcome to ViewerVoice, the tool that allows you to gain insights into what viewers have commented under a YouTube video. We created this dashboard particularly to help YouTubers better understand their audience and allow for data-driven decision making, enabling them to strategise and build their brand. As well as to help brands understand the audience of potential creators they are looking to partner with.
+
+This dashboard is still under development; further updates will be implemented in due course.  
+
+## 🚶‍♂️Walkthrough  
+### 🗨️Retrieving YouTube Comments  
+To retrieve comments we used the YouTube Data API which provides various functionalities, including retrieving YouTube comments. At present we do not read in comment replies. Please refer to the instructions provided in the following link to acquire your unique API key. Please be aware that each API key facilitates up to 10,000 API calls within a 24-hour period.  
+https://developers.google.com/youtube/v3/getting-started
+
+### 👩‍💻Implementing Natural Language Processing Models  
+ViewerVoice covers the following areas of NLP driven by BERT based models.
+
+- **Topic modelling** groups the comments into topics so that you can gain clearer insights into the key themes viewers are discussing in the comment section. For optimal performance of the current topic model, we recommend retrieving thousands of comments.
+
+- **Sentiment analysis** identifies whether comments are positive, negative or neutral. This provides you with an overview of what viewers feel toward aspects of the content. Please note that at present, the sentiment analysis does not take emojis into account.
+
+- **Semantic search** allows you to not only search for comments with exact matches to words, but also for comments that contain similar words. For example, if you search for 'music', comments containing 'music' as well as 'song' and 'vinyl' can show up. This feature enables you to search for specifics in a comment section - for example, perhaps you are a YouTuber that has collaborated with another creator or advertised for a brand you have partnered with, you can search for the name of the creator or brand in your comments to see how your viewers reacted to this.
+
+### 📊Creating a Streamlit WebApp
+Streamlit is an open-source Python library that enables the creation and sharing of data driven web applications. Leveraging Streamlit's versatile functions, we've tailored our dashboard to cover a range of essential features, such as:
+
+- Interactive input widgets: allowing users to query their preferred videos and apply personalised filters.
+- Comprehensive Python graphs: seamlessly integrating a variety of visualisations into the app.
+- Custom HTML and CSS: offering the freedom to fine-tune the application's style to our preferences.
+
+## 🔍Results  
+We hope the insights provided by ViewerVoice will enable YouTubers to cater to their audience and use data-derived observations to negotiate opportunities and aid brands in gaining a deeper understanding of the intended audience of prospective partnerships. Future developments look to include but are not limited to improving the UI/UX design and implementing LLMs to give enhanced and more comprehensible results.

app.py

@@ -0,0 +1,383 @@
+import os
+from git import Repo
+import streamlit as st    
+import time
+from PIL import Image
+import base64
+from transformers import pipeline
+import spacy
+import googleapiclient
+import numpy as np
+from sentence_transformers import SentenceTransformer
+from matplotlib import colormaps
+from matplotlib.colors import ListedColormap
+
+GITHUB_PAT = os.environ['GITHUB']
+SENTIMENT = os.environ['SENTIMENT']
+EMBEDDING = os.environ['EMBEDDING']
+
+if not os.path.exists('repo_directory'):
+    try:
+        Repo.clone_from(f'https://marcus-t-s:{GITHUB_PAT}@github.com/marcus-t-s/yt-comment-analyser.git', 'repo_directory'  ) 
+    except:
+        st.error("Error: Oops there's an issue on our end, please wait a moment and try again.")
+        st.stop()
+
+from repo_directory.all_utils import *
+
+
+# Streamlit configuration
+st.set_page_config(
+    page_title="ViewerVoice | YouTube Comment Analyser",
+    layout="wide",
+    page_icon=Image.open('page_icon.png')
+)
+
+
+# Define and load cached resources
+@st.cache_resource
+def load_models():
+    sentiment_pipeline = pipeline("sentiment-analysis", model=f'{SENTIMENT}')
+    embedding_model = SentenceTransformer(f'{EMBEDDING}')
+    spacy_nlp = spacy.load("en_core_web_sm")
+    add_custom_stopwords(spacy_nlp, {"bring", "know", "come"})
+    return sentiment_pipeline, embedding_model, spacy_nlp
+
+
+@st.cache_resource
+def load_colors_image():
+    mask = np.array(Image.open('youtube_icon.jpg'))
+    Reds = colormaps['Reds']
+    colors = ListedColormap(Reds(np.linspace(0.4, 0.8, 256)))
+    with open("viewervoice_logo_crop.png", "rb") as img_file:
+        logo_image = base64.b64encode(img_file.read()).decode("utf-8")
+    return mask, colors, logo_image
+
+
+sentiment_pipeline, embedding_model, spacy_nlp = load_models()
+mask, colors, logo_image = load_colors_image()
+
+
+# Hide line at the top and "made with streamlit" text
+hide_decoration_bar_style = """
+    <style>
+        header {visibility: hidden;}
+        footer {visibility: hidden;}
+    </style>
+"""
+st.markdown(hide_decoration_bar_style, unsafe_allow_html=True)
+
+main_page = st.container()
+
+if 'YouTubeParser' not in st.session_state:
+    st.session_state['YouTubeParser'] = YoutubeCommentParser()
+if 'comment_fig' not in st.session_state:
+    st.session_state["comment_fig"] = None
+    st.session_state["wordcloud_fig"] = None
+    st.session_state["topic_fig"] = None
+    st.session_state["sentiment_fig"] = None
+if 'rerun_button' not in st.session_state:
+    st.session_state['rerun_button'] = "INIT"
+if 'topic_filter' not in st.session_state:
+    st.session_state['topic_filter'] = False
+if 'sentiment_filter' not in st.session_state:
+    st.session_state['sentiment_filter'] = False
+if 'filter_state' not in st.session_state:
+    st.session_state['filter_state'] = "INIT"
+if 'video_link' not in st.session_state:
+    st.session_state["video_link"] = None
+if 'num_comments' not in st.session_state:
+    st.session_state['num_comments'] = None
+
+# Set reference to YouTubeParser object for more concise code
+yt_parser = st.session_state['YouTubeParser']
+
+
+def query_comments_button():
+    # Delete larger objects from session state to later replace
+    del st.session_state["comment_fig"]
+    del st.session_state["wordcloud_fig"]
+    del st.session_state["topic_fig"]
+    del st.session_state["sentiment_fig"]
+    del st.session_state["YouTubeParser"]
+
+    # Reset session state variables back to placeholder values
+    st.session_state.rerun_button = "QUERYING"
+    st.session_state['filter_state'] = "INIT"
+    st.session_state["topic_filter"] = False
+    st.session_state["sentiment_filter"] = False
+    st.session_state["semantic_filter"] = False
+    st.session_state["figures_built"] = False
+    st.session_state["comment_fig"] = None
+    st.session_state["wordcloud_fig"] = None
+    st.session_state["topic_fig"] = None
+    st.session_state["sentiment_fig"] = None
+    st.session_state["YouTubeParser"] = YoutubeCommentParser()
+
+
+def filter_visuals_button():
+    st.session_state["filter_state"] = "FILTERING"
+
+
+with st.sidebar:
+    st.session_state["api_key"] = st.text_input('YouTube API key', value="", type='password')
+    st.session_state["video_link"] = st.text_input('YouTube Video URL', value="")
+    st.session_state["max_comments"] = st.slider(label="Maximum number of comments to query",
+                                                 min_value=100,
+                                                 max_value=3000,
+                                                 step=100)
+    st.session_state["max_topics"] = st.slider(label="Maximum number of topics",
+                                               min_value=5,
+                                               max_value=20,
+                                               step=1)
+    st.button('Query comments :left_speech_bubble:', on_click=query_comments_button)
+
+with main_page:
+    # Reduce space at the top
+    reduce_header_height_style = """
+        <style>
+            div.block-container {padding-top:0rem;}
+            div.block-container {padding-bottom:1rem;}
+            div.block-container {padding-left:1.5rem;}
+        </style>
+    """
+    st.markdown(reduce_header_height_style, unsafe_allow_html=True)
+
+    # Title and intro section
+    markdown_content = f"""
+    <div style='display: flex; align-items: center; justify-content: center;'>
+        <img src='data:image/png;base64,{logo_image}' height='135px';/>
+    </div>
+    """
+    st.markdown(markdown_content, unsafe_allow_html=True)
+
+    # LinkedIn links
+    lnk = '<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/4.7.0/css/font-awesome.min.css">'
+    st.markdown(lnk + """
+    <div style="display: flex; justify-content: center; align-items: center; flex-direction: column;">
+        <br>
+        <p style="text-align: center;"><b>Made by</b>
+        <b>
+        <a href='https://www.linkedin.com/in/afiba-7715ab166/' style="text-decoration: none">
+        <i class='fa fa-linkedin-square'></i>&nbsp;<span style='color: #000000'>Afiba Annor</span></a>     
+        <a href='https://www.linkedin.com/in/marcus-singh-305927172/' style="text-decoration: none">
+        <i class='fa fa-linkedin-square'></i>&nbsp;<span style='color: #000000'>Marcus Singh</span></a>
+        </b></p>
+        </div>
+
+    """, unsafe_allow_html=True)
+
+    st.markdown("<hr>", unsafe_allow_html=True)
+    # Notes section
+    st.markdown("<p style='font-size: 1.3rem;'><b>📝 Notes</b></p>", unsafe_allow_html=True)
+
+    html_content = """
+    <ul>
+        <li style='font-size: 0.95rem;'>This dashboard is still under development; further updates will be implemented 
+        in due course.</li>
+        <li style='font-size: 0.95rem;'>Kindly refer to the instructions provided in this 
+        <a href='https://developers.google.com/youtube/v3/getting-started'>link</a>. 
+        This will guide you in acquiring your API key to retrieve comments.</li>
+        <li style='font-size: 0.95rem;'>Please be aware that each API key facilitates up to 10,000 API calls within 
+        a 24-hour period.</li>
+        <li style='font-size: 0.95rem;'>Currently, the dashboard caters to comments in English and does not 
+        include comment replies.</li>
+        <li style='font-size: 0.95rem;'>Comments undergo cleaning and pre-processing to optimise modelling. As a result, 
+        the returned comment count may fall short of the maximum queried amount.</li>
+        <li style='font-size: 0.95rem;'>Please note that the sentiment analysis currently does not take emojis into 
+        account.</li>
+        <li style='font-size: 0.95rem;'>For optimal performance of the current topic model, we recommend retrieving 
+        thousands of comments.</li>
+        <li style='font-size: 0.95rem;'>Please anticipate that querying comments and running the models may require 
+        a few minutes to complete.</li>
+    </ul>
+    <hr>
+    """
+    # Display the HTML content using st.markdown()
+    st.markdown(html_content, unsafe_allow_html=True)
+
+# Query comments section
+if (st.session_state.rerun_button == "QUERYING") and (st.session_state["video_link"] is not None):
+    with st.spinner('Querying comments and running models'):
+        yt_parser = st.session_state["YouTubeParser"]
+        try:
+            yt_parser.build_youtube_api(st.session_state['api_key'])
+        except:
+            st.error("Error: Unable to query comments, please check your API key")
+            st.stop()
+        try:
+            yt_parser.query_comments(st.session_state['video_link'], st.session_state['max_comments'])
+        except googleapiclient.errors.HttpError:
+            st.error("Error: Unable to query comments, please check your API key.")
+            st.stop()
+        except:
+            st.error("Error: Unable to query comments, incorrect YouTube URL or maximum \
+                              API call limit reached.")
+            st.stop()
+
+        # Run formatting and models
+        yt_parser.format_comments()
+        yt_parser.clean_comments()
+        yt_parser.run_sentiment_pipeline(sentiment_pipeline)
+        yt_parser.run_topic_modelling_pipeline(embedding_model,
+                                               nlp=spacy_nlp,
+                                               max_topics=st.session_state['max_topics'])
+    # Set "QUERY COMPLETE" to bypass running this section on script re-run
+    st.session_state.rerun_button = "QUERY COMPLETE"
+
+# Once comments are queried, build charts ready to visualise
+if st.session_state.rerun_button == "QUERY COMPLETE":
+    # Check for built figures:
+    if (not st.session_state["figures_built"]) or (st.session_state.filter_state == "FILTERING"):
+        # Select colors for wordcloud
+
+        # If filtering button pressed
+        if st.session_state.filter_state == "FILTERING":
+            df_filtered = yt_parser.df_comments.copy()
+            if st.session_state["topic_filter"]:
+                df_filtered = df_filtered.query(f"Topic == {st.session_state.topic_filter}")
+            if st.session_state["sentiment_filter"]:
+                df_filtered = df_filtered.query(f"Sentiment == {st.session_state.sentiment_filter}")
+            if st.session_state["semantic_filter"]:
+                df_filtered = semantic_search(df=df_filtered, query=st.session_state["semantic_filter"],
+                                              embedding_model=embedding_model,
+                                              text_col='Comment_Clean')
+            if len(df_filtered) == 0:
+                st.session_state['num_comments'] = 0
+
+            else:
+                st.session_state['num_comments'] = len(df_filtered)
+                # Build filtered table figure
+                st.session_state["table_fig"] = comments_table(df_filtered,
+                                                               ['publishedAt', 'Comment_Formatted', 'Likes',
+                                                                'Sentiment', 'Topic'],
+                                                               {'publishedAt': 'Date', 'Comment_Formatted': 'Comment'})
+
+                # Build filtered wordcloud figure
+                st.session_state["wordcloud_fig"] = comment_wordcloud(df_filtered, mask, colors)
+
+                # Build filtered topic figure
+                st.session_state["topic_fig"] = topic_treemap(df_filtered, "Topic")
+
+                # Build filtered sentiment figure
+                st.session_state["sentiment_fig"] = sentiment_chart(df_filtered, "Sentiment")
+
+                st.session_state["figures_built"] = True
+
+                st.session_state.filter_state = "FILTERED"
+
+        # No filtering selected
+        else:
+            st.session_state['num_comments'] = len(yt_parser.df_comments)
+
+            # Can only build graphs if we have comments
+            if st.session_state['num_comments'] > 0:
+                try:
+                    # Build unfiltered table figure
+                    st.session_state["table_fig"] = comments_table(yt_parser.df_comments,
+                                                                   ['publishedAt', 'Comment_Formatted', 'Likes',
+                                                                    'Sentiment', 'Topic'],
+                                                                   {'publishedAt': 'Date',
+                                                                    'Comment_Formatted': 'Comment'})
+                    # Build unfiltered wordcloud figure
+                    st.session_state["wordcloud_fig"] = comment_wordcloud(yt_parser.df_comments,
+                                                                          mask, colors)
+                    # Build unfiltered topic figure
+                    st.session_state["topic_fig"] = topic_treemap(yt_parser.df_comments, "Topic")
+                    # Build unfiltered sentiment figure
+                    st.session_state["sentiment_fig"] = sentiment_chart(yt_parser.df_comments, "Sentiment")
+
+                    st.session_state["figures_built"] = True
+                except:
+                    st.error("Error: Oops there's an issue on our end, please wait a moment and try again.")
+                    st.stop()
+
+with main_page:
+    if st.session_state.rerun_button == "QUERY COMPLETE":
+        st.subheader(f"{yt_parser.title}")
+        st.markdown("<hr><br>", unsafe_allow_html=True)
+
+        if st.session_state['num_comments'] > 0:
+            table_col, word_cloud_col = st.columns([0.55, 0.45])
+            with table_col:
+                st.markdown(f"""<p style='font-size: 1.3rem;
+                display: flex; align-items: center; justify-content: center;'><b>
+                                                    Comments</b></p>""", unsafe_allow_html=True)
+                st.plotly_chart(st.session_state["table_fig"], use_container_width=True)
+
+            with word_cloud_col:
+                st.markdown(f"""<p style='font-size: 1.3rem;
+                display: flex; align-items: center; justify-content: center;'><b>
+                                                               Word Cloud</b></p>""", unsafe_allow_html=True)
+
+                st.pyplot(st.session_state["wordcloud_fig"], use_container_width=True)
+
+            treemap_col, sentiment_donut_col = st.columns([0.55, 0.45])
+
+            with treemap_col:
+                st.markdown(f"""<p style='font-size: 1.3rem;
+                display: flex; align-items: center; justify-content: center;'><b>
+                                                                Topic Proportions</b></p>""", unsafe_allow_html=True)
+
+                st.plotly_chart(st.session_state["topic_fig"], use_container_width=True)
+
+            with sentiment_donut_col:
+                st.markdown(f"""<p style='font-size: 1.3rem;
+                display: flex; align-items: center; justify-content: center;'><b>
+                                                        Sentiment Distribution</b></p>""", unsafe_allow_html=True)
+                st.plotly_chart(st.session_state["sentiment_fig"], use_container_width=True)
+
+            # st.table(yt_parser.df_comments.head())
+        else:
+            st.write("Unfortunately we couldn't find any comments for this set of filters, please try "
+                     "editing the filters and try again")
+
+with st.sidebar:
+    # Define the HTML and CSS for the button-style container
+    if st.session_state['num_comments'] is not None:
+        num_comments = st.session_state['num_comments']
+    else:
+        num_comments = 0
+    htmlstr = f"""
+                <p style='background-color: rgb(255, 255, 255, 0.75); 
+                color: rgb(0, 0, 0, 0.75); 
+                font-size: 40px;    
+                border-radius: 7px; 
+                padding-top: 25px; 
+                padding-bottom: 25px;
+                padding-right: 25px; 
+                padding-left: 25px; 
+                line-height:25px;
+                display: flex;
+                align-items: center;
+                justify-content: center;
+                box-shadow: 0 0 10px rgba(0, 0, 0, 0.1);'>
+                &nbsp;{num_comments}</p>
+                """
+    # Display the button-style container with number of comments
+    st.subheader("Number of comments")
+    st.markdown(htmlstr, unsafe_allow_html=True)
+
+    # Filters section
+    st.subheader("Filters")
+
+    if yt_parser.df_comments is not None:
+        st.session_state["topic_filter"] = st.multiselect("Topic",
+                                                          options=sorted(list(yt_parser.df_comments['Topic'].unique())))
+        st.session_state["sentiment_filter"] = st.multiselect("Sentiment",
+                                                              options=list(yt_parser.df_comments['Sentiment'].unique()))
+        st.session_state["semantic_filter"] = st.text_input("Keyword search",
+                                                            max_chars=30)
+        st.button('Filter visualisations :sleuth_or_spy:', on_click=filter_visuals_button)
+
+    else:
+        st.multiselect("Topic",
+                       options=["Please query comments from a video"],
+                       disabled=True)
+        st.multiselect("Sentiment",
+                       options=["Please query comments from a video"],
+                       disabled=True)
+        st.text_input("Keyword search",
+                      disabled=True)
+        st.button('Please query comments before filtering',
+                  disabled=True)

requirements.txt

@@ -0,0 +1,20 @@
+streamlit
+numpy==1.24.4
+urllib3==1.26.16
+beautifulsoup4==4.12.2
+git+https://github.com/scikit-learn-contrib/hdbscan.git
+bertopic==0.15.0
+contractions==0.1.73
+google_api_python_client==2.96.0
+pandas==1.5.3
+python-dotenv==1.0.0
+scikit_learn==1.2.2
+sentence_transformers==2.2.2
+spacy==3.6.1
+https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.6.0/en_core_web_sm-3.6.0.tar.gz
+torch==2.0.1
+transformers==4.31.0
+umap==0.1.1
+umap_learn==0.5.3
+wordcloud==1.9.2
+#need to install C++ this for hbscan https://visualstudio.microsoft.com/visual-cpp-build-tools/