Next Gen Recommendation Systems

Welcome to the Next Gen Recommendation Systems project! In this repository, we developed three distinct recommendation systems, each exploring a different approach and catering to unique use cases. Here’s a brief overview of the project structure, methods, and goals.

---

Project Overview

1. Cross-Domain Recommendation System (Movies & Books)
2. Generative AI-Based Personalized Recommendation System
3. Session-Based Recommendation System

Each system serves a unique purpose and is built with different data-processing pipelines, machine learning models, and evaluation metrics. Additionally, we provide a responsive web application, FlexRead, for user interaction with these recommendations.

---

1. Cross-Domain Recommendation System (Movies & Books)

This system aims to bridge recommendations across two domains—books and movies—by creating a genre-based "bridge" that allows us to recommend items in one domain (e.g., movies) based on items from another (e.g., books).

• Custom Genre Mapping: To connect books and movies, we categorized both into custom genre types that capture similar thematic elements across domains.
• Modeling Approach: We used a RoBERTa model for text-based embedding of content and a K-Nearest Neighbors (KNN) approach to recommend items within these custom genres.
• FlixRead: Users can view, filter, and interact with recommendations through the FlexRead webpage, which is designed to be responsive and user-friendly.
• 

2. Generative AI-Based Personalized Recommendation System

With the rise of Generative AI, we implemented a recommendation system focused on personalized recommendations, such as daily meal suggestions based on user preferences and needs.

• Personalization Factors: Recommendations consider user location, cuisine preferences, lifestyle, and health requirements to deliver highly personalized suggestions.
• Model Evaluation: We aim to assess the accuracy and relevancy of AI models like Gemini and Ollama by comparing generated recommendations with real-life user data.
• Future Directions: Ongoing research includes measuring the accuracy and personalization of these AI models and enhancing our models’ responsiveness to user-specific needs.

3. Session-Based Recommendation System

The goal of this recommendation system is to suggest the next item based on a user’s current session and past activities, such as clicks, cart additions, or purchases.

• Modeling Approach: Built using SR-GNN (Session-Based Graph Neural Network) and PyGeometry, we designed a data pipeline to preprocess session data and train a model for next-item recommendations.
• Performance Metrics: Currently, the system achieves a hit rate of 4 out of 10 and an accuracy rate of 20-30% in predicting relevant items.
• Applications: This system is valuable for e-commerce, where real-time session data can drive personalized recommendations.

---

Directory Structure

.
├── cross_domain_recommendation/
│   ├── genre_mapping/
│   ├── models/
│   └── roberta_knn.py
├── gen_ai_recommendation/
│   ├── gemini_eval/
│   ├── ollama_eval/
│   └── personalized_meal_recommender.py
├── session_based_recommendation/
│   ├── data_pipeline/
│   ├── sr_gnn_model/
│   └── session_recommender.py
├── flexread/
│   ├── static/
│   ├── templates/
│   └── app.py
└── README.md

---

Future Enhancements

• Cross-Domain System: Expand to additional content domains (e.g., music, podcasts).
• Generative AI System: Extend AI-based recommendations to broader lifestyle suggestions (e.g., workout plans).
• Session-Based System: Improve accuracy and explore real-time feedback integration for instant recommendations.

Thank you for exploring our Next Gen Recommendation Systems!