I Built a Smart Movie Recommender with Collaborative Filtering

2026-02-14 · Source: Analytics Vidhya · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, medium

Summary

This article details the construction of a movie recommendation system utilizing collaborative filtering and Singular Value Decomposition (SVD), enhanced with OpenAI's GPT-4.1-mini and text-embedding-3-small models. The system addresses common downsides of traditional user-user and item-item filtering, such as computational slowness and data sparsity. It employs The Movies Dataset, specifically the "ratings_small.csv" file containing 100,000 ratings from 700 users on 9,000 movies, for training. The Surprise library is used for data splitting and SVD implementation, with `GridSearchCV` optimizing SVD parameters to achieve a best RMSE score of 0.890276. A unique "smart twist" integrates an LLM to generate a question for the user based on five SVD-predicted movie recommendations, then uses cosine similarity between the user's response and movie descriptions to provide a final, mood-based recommendation.

Key takeaway

For Data Scientists building recommendation systems, integrating Large Language Models (LLMs) with traditional collaborative filtering techniques like SVD can significantly enhance personalization. Your systems can move beyond static predictions to offer mood-based recommendations, improving user engagement. Consider using the Surprise library for efficient SVD implementation and OpenAI's APIs for dynamic user interaction and semantic matching to refine your recommendation logic.

Key insights

SVD-based collaborative filtering combined with LLMs offers dynamic, mood-aware recommendations.

Principles

• Matrix factorization mitigates data sparsity.
• Grid search optimizes model parameters.
• LLMs can personalize user interaction.

Method

The method involves using the Surprise library for SVD-based collaborative filtering, optimizing parameters with `GridSearchCV`, and then integrating an OpenAI LLM to generate a user-specific question for final recommendation selection via cosine similarity.

In practice

• Use Surprise library for recommendation models.
• Employ `GridSearchCV` for SVD parameter tuning.
• Integrate LLMs for interactive, mood-based filtering.

Topics

• Recommendation Systems
• Collaborative Filtering
• Singular Value Decomposition
• Large Language Models
• OpenAI API Integration

Best for: Machine Learning Engineer, Data Scientist, AI Student

Related on AIssential

• AgenticRAG vs Traditional RAG — LLMs can dynamically extract features from natural text to enrich user profiles and personalize recommendations.
• Understanding Stability in Modern Vector Databases, A Generative Paradigm Shift for Click-Through Rate Prediction, and More! — Modern vector retrieval and recommendation systems overcome dimensionality challenges and scale through specialized architectures and efficient processing.
• Paper Digest: CIKM 2025 Papers & Highlights — CIKM 2025 research highlights advancements in LLMs, graph learning, and multimodal data, addressing challenges in recommendation, forecasting, and privacy.
• 2025 Year in Review — Integrating LLMs with RecSys can yield powerful, steerable recommendation systems capable of cross-platform suggestions.
• Rethinking Personalization for the Agent Era, Semantic Recall for Vector Search, and More! — LLMs are increasingly central to advancing information retrieval and recommender system capabilities.
• Personalization in the Era of LLMs - Shivam Verma, Spotify — Spotify is integrating LLMs into its recommendation system using user embeddings, semantic IDs, and soft tokens for enhanced personalization.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Analytics Vidhya.