Sistema recomendador basado en grafos y en información textual: Un enfoque híbrido para generar recomendaciones explicables.

Abstract

Abstract: The present work proposes a hybrid recommender system that integrates knowledge graphs and textual information, using generative language models, with the aim of providing precise and explainable recommendations. To this end, a systematic review of the literature was carried out where the most relevant approaches used in graph-based recommendation systems and textual knowledge were identified today. Consequently, an architecture based on GraphRAG was designed that combines structured recovery with natural language generation, which allowed to offer recommendations supported by explicit and traceable evidence. The system integrates repositories of structured and unstructured data retrieved from Scopus into a knowledge graph modeled in Neo4j, as well as a conversational module capable of interpreting queries made in natural language. The prototype, once tested, empirically validated its viability through a multidimensional evaluation, reaching high levels of accuracy, relevance and fidelity. As observed in the results, the convergence between graphs and generative models significantly improves the quality, transparency and reliability of the recommendations.