ViLLA-MMBench Documentation

Overview

ViLLA-MMBench is a unified benchmark suite for LLM-Augmented Multimodal Movie Recommendation. It provides a comprehensive framework for experimenting with different recommendation models using visual, audio, and textual modalities.

Check ViLLA-MMBench on GitHub
Check How to use ViLLA-MMBench in Google Colab
Check all the Framework's in Google Colab (General Test)

Project Structure

• villa_mmbench/ - Main package directory
  • data/ - Data loading and processing modules
  • process/ - Core processing and model implementation
  • utils/ - Utility functions and helpers
• colabs/ - Jupyter notebooks for various analyses
• outputs/ - Generated output files
• reclist/ - Recommendation list outputs

Getting Started

# Clone the repository
git clone git@github.com:RecSys-lab/ViLLA-MMBench.git

# Create and activate virtual environment
python -m venv venv
.\venv\Scripts\activate  # Windows
source venv/bin/activate  # Linux/Mac

# Install package
pip install -e .

Main Components

Data Processing

The framework includes modules for processing different modalities:

• Text Processing (LLM-based)
• Visual Features (AVF, CNN)
• Audio Features (BLF, i-vector)

Models

Supported recommendation models include:

• Collaborative Filtering (CF)
• Visual Bayesian Personalized Ranking (VBPR)
• Adversarial Multimedia Recommendation (AMR)
• Visual Matrix Factorization (VMF)
• Variational Autoencoder CF (VAECF)