🤖 RoboReviews

Overview

This repository contains the source code, documentation, and deliverables for a Video Game & Reviews ML/AI project. The goal is to process and analyze video game data and reviews. It performs sentiment classification, clusters game genres, and generates summaries to recommend top games.

🎮 Features

1. 🔍 Sentiment Analysis:

   • Classifies reviews as Positive, Neutral, or Negative.
   • Fine-tuned DistilBERT ensures high accuracy tailored to the dataset.

2. 🗂️ Category Clustering:

   • Groups games into broader genre categories, such as Combat-Focused Gameplay, etc..
   • Enables better data organization and visualization.

3. 📝 Review Summarization:

   • Generates blog-like articles summarizing game features.
   • Highlights the top three games per cluster and reasons why people like/dislike them.

4. 🌐 Interactive Website:

   • Presents all analyses in an intuitive, user-friendly interface.
   • Allows live sentiment processing of user review texts.

📊 Datasets

• Primary Dataset: Custom video game review dataset.
  • UCSD Amazon Reviews

🏗️ Project Structure

root/
├── core/               # Core Django app
│   └── templates/      # HTML index view
├── data/               # Raw and processed datasets
├── models/             # Saved and fine-tuned models
│   ├── clustering/     # pyLDAvis visualization
│   └── sentiment/      # DistilBERT files
├── notebooks/          # Jupyter notebooks for model development
├── scripts/            # Python scripts for feeding the database
├── served_model/       # Flask app serving TinyBERT
├── static/             # Static files (CSS, JS, images)
├── db.sqlite3          # Django SQLite database
├── manage.py           # Django CLI utility script
├── README.md           # Project documentation
└── requirements.txt    # Python dependencies

⚙️ How It Works

1. 🧹 Preprocessing:

   • Text Cleaning: Removes special characters and standardizes text.
   • Data Cleaning: Drops unnecessary columns and handles missing values.
   • Enrichment: Adds genres via the OpenAI API.
   • Balancing: Applies upsampling and downsampling.
   • Normalization: Performs lemmatization and stopword removal.
   • Tokenization & Vectorization: Prepares text for modeling.

2. 📈 Model Pipeline:

   • Sentiment Classification: Uses fine-tuned DistilBERT for sentiment analysis.
   • Topic Modeling: Employs LDA to uncover hidden topics and group similar game genres.
   • Summarization: Utilizes the OpenAI API for concise summaries of game pros and cons.

3. 📊 Evaluation:

   • Metrics: Evaluates model performance using accuracy, precision, recall, and F1-score.
   • Visualization: Includes confusion matrix, word clouds, and t-SNE plot.
   • Analysis: Displays example predictions and enables interactive topic exploration with pyLDAvis.

4. 🚀 Deployment:

   • Web Interface: Built with Django.
   • Model Serving: Sentiment model served via Flask.
   • Hosting: Entire application hosted on Heroku.

📦 Deliverables

1. 📜 Source Code:

   • Organized Python scripts and Jupyter notebooks.

2. 🌐 Website:

   • Live demo hosted (RoboReviews).

3. 📊 Evaluation Metrics:

   • Visualizations: Plots (images/notebooks) and LDA visualization rendered as HTML.

💡 Usage

• 💬 Sentiment Predictions: Users can test written texts for sentiment.
• 📊 Review Analysis: View categorized and summarized results.

🚧 Future Enhancements

• Extend datasets for broader coverage.
• Fine-tune and host LLM for game summarization.

🙌 Acknowledgments

• 📚 Datasets from UCSD.
• 🛠️ Pretrained models from Hugging Face.