Sistema de recomendación personalizado mediante LLMs.

Abstract

Abstract: This work focuses on developing a personalized recommendation system based on collaborative filtering approach using deep neural networks and user and item interactions and characteristics. A preprocessed MovieLens 1M dataset is used. The goal is to predict user preferences and offer justifications using a Large Language Model. We implemented a hybrid model called DeepHybrid, which fuses collaborative embeddings with precomputed textual embeddings into a deep neural network for accurate rating regression. The workflow included data analysis and light preprocessing to prepare the model inputs. Performance was evaluated using MAE, MSE, RMSE, and R², proving the model's competitiveness against simpler baselines. To increase transparency, we added an explainability module based on an LLM. This module generates natural language explanations based on the user's profile, history, and specific recommendations, thereby improving the end user's understanding of the system's output.