yunqili4/cs410-final-project

CS410 Final Project -- Amazon Review Summarization and Sentiment Analysis

Dataset

Available here, we use four categories, All_Beauty, Digital_Music, Handmade_Product, and Health_and_Personal_Care.

Workflow

Data preprocessing -> Sentiment classification (group positive and negative reviews to proceed) -> Fine-tuning summarization model on training data -> Evaluate summarization model on test data.

Models

• Sentiment classification uses pre-trained DistillBERT.
• Review summarization uses facebook/bart-large-cnn fine-tuned on category of review dataset.

Layout

• checkpoints folder contains fine-tuned models for each specific categories of dataset.
• src folder contains source code.
• docs records experiments results.

Usage

Run sentiment classification

python src/classification.py [category]

Run fine-tuning

python src/finetune.py [category]

Run summarization, you should firstly obtain an Anthropic Claude API key, and

export ANTHROPIC_API_KEY='your-api-key-here'

then

python src/summarization.py