This project involves preparing, analyzing, and fine-tuning dataset McAuley-Lab/Amazon-Reviews-2023 for product reviews. (The finetuned model can be downloded from this link)
It includes data preparation, model fine-tuning, user interaction for review summarization, and sentiment analysis.
Below are the detailed steps and functionalities:
Run the data_save.py script to download the dataset.
Match the product names in the metadata using the ASIN code.
Remove any unnecessary details to clean the data.
The train.py file is used to fine-tune the BART-large-cnn model for generating summaries of product reviews.
Use the user_input.ipynb notebook to:
- Select a product from a provided list.
- Obtain a summarized review of the selected product.
- View the sentiment analysis of the selected product.
The Comapring_Roberta_and_Vader_models.ipynb notebook is used to compare two models for sentiment analysis:
Here, the comparison is done for a subset of first 250 data samples from the McAuley-Lab/Amazon-Reviews-2023 dataset.
- RoBERTa Model: RoBERTa (Robustly optimized BERT approach) is an advanced transformer-based model designed for natural language understanding. It builds upon the BERT (Bidirectional Encoder Representations from Transformers) architecture
- VADER Model: VADER (Valence Aware Dictionary and sEntiment Reasoner) is a rule-based model specifically attuned to sentiment analysis, particularly suited for social media texts. It is designed to be fast and computationally efficient, making it ideal for real-time applications
Both RoBERTa and VADER have their strengths and are suited for different scenarios. RoBERTa, with its transformer-based architecture, excels in deep contextual understanding and complex text analysis. VADER, with its rule-based approach, provides fast and interpretable results but it has limitations to capture wider range of sentiments.