A machine-learning project that classifies Amazon product reviews as real or fake based on their text content.

This project reads Amazon reviews from CSV data, cleans the text, engineers features, trains multiple classifiers, and serves predictions through a Streamlit dashboard. The workflow covers data loading, feature engineering, model training/evaluation, and an interactive frontend for live inference.

Amazon-Review-Analyzer-2/
├── data/            # Raw and processed review data files
├── model/           # Trained joblib models, metadata, and BERT LoRA adapter files
├── src/             # Python scripts for data processing, training, and evaluation
├── webapp/          # Frontend application for serving model predictions
├── pyproject.toml   # Project metadata and dependencies (used by uv)
├── .gitignore       # Files and directories excluded from version control
└── README.md        # Project documentation

This project uses uv for Python project and environment management.

macOS / Linux

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

After installation, restart your terminal so the uv command is available.

1. Clone the repository

   git clone https://github.com/wijayaju/Amazon-Review-Analyzer-2.git
   cd Amazon-Review-Analyzer-2

2. Create a virtual environment and install dependencies

   uv sync

   This reads pyproject.toml, creates a .venv virtual environment, and installs all listed dependencies.

3. Add new packages as needed

   uv add <package-name>

4. Run scripts

   uv run python src/<script>.py

1. Place the raw review data file in the data/ directory.

2. Run preprocessing to create the engineered dataset:

   uv run python src/preprocess.py --input "data/<INPUT_CSV_PATH>.csv"

   Example:

   uv run python src/preprocess.py --input "data/fake-reviews.csv"

   Output is written to data/preprocessed_reviews.csv.

3. Use the scripts in src/ to train the baseline TF-IDF + logistic regression model, the XGBoost model, or the BERT LoRA model.

4. Trained artifacts are saved in model/ as joblib files, JSON metadata, and a bert_lora/ adapter directory.

5. Launch the web application from webapp/ to interact with the model through a browser.

Start an interactive UI where you can paste a review and choose a model (baseline, xgboost, or bert) for instant classification:

uv run streamlit run webapp/streamlit_app.py

The app predicts whether the review is AI-generated or human-written, displays model confidence, and shows extracted feature values for the current input.

I used GitHub Copilot to help draft and scaffold parts of this project. I am the one responsible for reviewing, testing, and revising any AI-generated output before treating it as a final product. AI assistance is primarily used to accelerate development, but is not used to replace my own judgment.