Amazon Product Query Assistant

This project explores how to build a smart Amazon product query assistant using multiple information retrieval and generation methods. It compares BM25 keyword retrieval, semantic search with embeddings, and a Hybrid RAG pipeline on the Amazon Reviews 2023 dataset. The system supports both retrieval-only search and retrieval-augmented generation (RAG), and presents the results through an interactive Streamlit app where users can explore product results or receive LLM-generated answers based on retrieved Amazon product metadata and reviews.

The web interface can now be accessed at https://yhouyang02-dsci-575-project-yuhengo-mkcchoy.share.connect.posit.cloud.

Project maintainers

• Mickey Choy
• Yuheng Ouyang

Repository structure

DSCI_575_project_yuhengo_mkcchoy
  ├── app/                     # Streamlit app code
  ├── bm25_index/              # BM25 retriever artifacts
  ├── data/                    # Raw and processed data (downloaded separately)
  ├── notebooks/               # Jupyter notebooks for experimentation
  ├── results/                 # Result discussion and analysis
  ├── semantic_index/          # Semantic retriever artifacts
  ├── src/                     # Source code for data processing and retrieval
  ├── environment.yml          # Conda environment specification
  ├── README.md                # Project overview and instructions

Get started

To run the app locally, follow the following steps:

1. Clone the repository and navigate to the project directory.

   git clone https://github.com/UBC-MDS/DSCI_575_project_yuhengo_mkcchoy.git
   cd DSCI_575_project_yuhengo_mkcchoy

2. Create and activate the conda` environment.

   conda env create -f environment.yml
   conda activate dsci-575-mkc-yho

3. To use the RAG mode, create a Hugging Face Access Token following their instruction. When creating a new token,

   • Choose "Fine-grained" for token type
   • Check all permissions under "Repositories" and "Inference"

4. Add the API token into your local folder. This should not be committed to any remote repositories. Replace <your-api-token> with the token value you just created and run the following in your terminal. You must have this ready to use the LLM-powered RAG mode.

   echo "HUGGINGFACEHUB_API_TOKEN=<your-api-token>" > .env

5. Start the Streamlit app. The app should open in your default web browser at http://localhost:8501. If it does not open automatically, you can navigate to that URL manually. It can take a few minutes to load the full app.

   streamlit run app/app.py

6. Enter product-related queries in the input box and click the "Search" button. The results may be limited since our test models are built on a subset of the full dataset. For better results, you can try queries related to the "Appliances" category, such as "quiet dishwasher stainless steel".

7. To stop the app, press Ctrl+C in the terminal where the Streamlit app is running.

8. For developers who want to retrain the models with a different sample size or adjust the training process, you can modify src/build_artifacts.py and run the following command to rebuild the model artifacts. This will retrain the BM25 and semantic retrievers and save the new artifacts in their respective folders.

   python src/build_artifacts.py

RAG workflow diagram

flowchart LR
    A[User Query] --> B{Retriever}
    B --> C[Semantic Retriever<br/>FAISS + Embeddings]
    B --> D[Hybrid Retriever<br/>BM25 + Semantic + RRF]

    C --> E[Top-k Documents]
    D --> E

    E --> F[build_context]
    F --> G[Prompt Template]
    G --> H[Hugging Face LLM]
    H --> I[Generated Answer]

    J[Amazon product docs<br/>reviews + metadata] --> C
    J --> D
    K[.env API key] --> H

Loading