📰 Topic Modeling & Aspect-Based Sentiment Analysis of BBC News

🚀 Analyzing political news from the BBC using Topic Modeling and Sentiment Analysis.
We use BERTopic with all-MiniLM-L12-v2 embeddings from HuggingFace to uncover hidden topics and extract insights from the text.

---

📌 Project Overview

• Data Source: BBC News articles from BigQuery (bigquery-public-data)

• Models Used: BERTopic, all-MiniLM-L12-v2, facebook/bart-large-cnn, gemini-2.0-flash

• Visualization: Interactive topic visualizations (barcharts, clusters, heatmaps)

• Text Summarisation:
  - summarised very long articles to avoid BERTopic bias with facebook/bart-large-cnn - summarised articles with facebook/bart-large-cnn & gemini-2.0-flash for comparison

  • Topic-Level Summarisation: labelled topics with gemini-2.0-flash

• Named Entity Recognition extraction of entities with RoBERTa and spaCy

• Sentiment Analysis: Aspect-based sentiment evaluation of political entities where aspect labels = topic labels

• Reference dataset:

• Sentiment Labels: Positive, Negative, Neutral

---

⚡ Quick Setup

pip install -r requirements.txt

Get Topics & Summaries by BART

cd src/
python main.py

---

Get Summaries by Gemini

cd src/
python summary_llm.py

Get Named Entities

pip install -r ner_requirements.txt
cd src/
python entities.py