MoviesRec: An Intelligent Conversational Movie Recommendation System Integrating Large Language Models and Collaborative Filtering

MoviesRec is an innovative intelligent conversational movie recommendation system whose core lies in integrating the reasoning and conversational capabilities of large language models (LLMs) with the precision of traditional collaborative filtering algorithms. Through LangGraph workflow orchestration, it achieves end-to-end automated recommendations including natural language understanding, dynamic SQL filtering, SVD collaborative filtering ranking, diversity re-ranking, and intelligent summarization, addressing the pain points of traditional recommendation systems such as lack of deep intent understanding and natural interaction.

In the era of information explosion, traditional recommendation systems rely on historical ratings for mechanical recommendations, lacking deep understanding of user intent and natural interaction capabilities. As an open-source system, MoviesRec adopts a "hybrid architecture" concept: it not only compensates for the limitations of pure LLMs in personalized recommendation accuracy but also solves the problem of traditional collaborative filtering lacking natural conversational capabilities. Through LangGraph, it builds an intelligent system that combines deep understanding and precise recommendation.

MoviesRec is based on a state-driven computation graph and includes five key nodes:

1. Intent Analysis Node: Parses user natural language input and extracts key preferences such as genre and style;
2. SQL Filtering Node: Dynamically generates SQL queries to filter candidate movies from the metadata database;
3. Collaborative Filtering Node: Uses the SVD model from scikit-surprise to predict and rank based on user-item rating data;
4. Diversity Re-ranking Node: Optimizes recommendation diversity through cluster analysis to avoid homogenization;
5. Summary Generation Node: Uses LLMs to convert results into natural conversational responses, enhancing interaction friendliness.

The technology selection includes:

• Backend: FastAPI + Uvicorn to build high-performance APIs, supporting asynchronous operations and Swagger documentation;
• LLM Orchestration: LangChain + LangGraph to implement workflows, with DeepAgents providing agent capabilities;
• Machine Learning: scikit-surprise (SVD), scikit-learn (clustering), Pandas/NumPy for data processing;
• Data Management: SQLite for metadata storage, JSON for session state persistence;
• Model Training: Offline script svd_model_trainer.py to train and save models, svd_model_predictor.py for real-time prediction.

The core advantages of MoviesRec include:

• Natural Interaction: Supports daily language to describe needs and proactively clarifies and guides;
• Precise Recommendation: Collaborative filtering is based on real user behavior, avoiding LLM "hallucinations";
• Interpretability: Recommendation processes are transparent, with explainable recommendation reasons;
• Scalability: Modular design supports replacing algorithms or LLM providers.

Deployment steps:

1. Configure the OpenAI API key;
2. Prepare movie datasets (movies.csv, ratings.csv, etc.);
3. Run the SVD model training script;
4. Start the FastAPI service (default listens on port 8000);
5. Test endpoints via Swagger UI.

MoviesRec represents an important direction for recommendation systems—the integration of traditional algorithms and emerging AI capabilities: LLMs are responsible for "understanding" and "expression", while collaborative filtering handles "precise recommendation", complementing each other. For AI developers, it provides a reference architecture for workflow design, state management, and balancing efficiency and quality. In the future, hybrid architectures are expected to be applied in more vertical fields.