Skip to content

Latest commit

 

History

History
68 lines (43 loc) · 2.47 KB

README.md

File metadata and controls

68 lines (43 loc) · 2.47 KB

nexREC: Graph-Based Movie Recommendation Engine 🎬

The Story Behind nexREC

nexREC was born out of a final-year academic project that took us on an incredible journey through the world of recommendation systems. What started as a simple idea evolved into a sophisticated graph-based recommendation engine that truly understands user preferences.

The Journey

The path to building nexREC wasn't straightforward. We began by exploring traditional recommendation systems, diving deep into content-based filtering and collaborative filtering. However, we quickly realized that user preferences are incredibly diverse and complex - influenced by social circles, trends, and personal experiences.

This realization led us to explore Graph Neural Networks (GNNs), a less-researched but promising area that could capture these complex relationships. After countless sleepless nights, hundreds of research papers, and numerous experiments, we developed a system that could instinctively respond to user needs.

Technical Implementation

Core Technologies

  • Backend Framework: Django & Django Rest Framework
  • Frontend: Remix
  • Vector Database: Milvus
  • Relational Database: PostgreSQL
  • Machine Learning:
    • Sentence Transformers (SBERT)
    • PyTorch
    • PyTorch Geometric
  • Deployment:
    • AWS Lightsail (Backend & Milvus)
    • Aiven Cloud (PostgreSQL)
    • HuggingFace (Models)
    • Vercel (Frontend)

Model Architecture

We experimented with various GNN architectures including:

  • SAGEConv
  • GAT (Graph Attention Networks)
  • GATv2

After extensive hyperparameter tuning and optimization, we achieved an RMSE loss between 0.94 and 1.21, which is quite solid considering real-world rating deviations and data skewness.

Data Processing

The project involved processing approximately 10k movies with user reviews. We used PySpark for efficient data cleaning and processing, which was a significant challenge in itself. The data distribution analysis and sampling experiments were crucial in achieving our final results.

Team

This project wouldn't have been possible without the incredible teamwork of:

  • Basanta Shrestha
  • Pawan Shah
  • Suwash Shrestha

Getting Started

Prerequisites

  • Python 3.8+
  • PostgreSQL
  • Milvus

Acknowledgments

Special thanks to Sushan Kattel for the invaluable guidance in PySpark and data processing.

Note: The deployed version is currently offline due to AWS billing considerations. We're working on making it available again soon!