Sistema de Recomendación de Recursos Educativos Basado en Learning Analytics, LLMs y Grafos del Conocimiento

Abstract

Abstract: The Curricular Integration Work (TIC) presents the design, development, and evaluation of a prototype educational recommendation system that integrates Knowledge Graphs (KG), Large Language Models (LLMs), and Learning Analytics (LA) to personalize educational resources. The system addresses the limitation of generalized virtual learning environments and proposes a hybrid approach (GraphRAG) that allows users to formulate queries in natural language, such that these are automatically translated into SPARQL queries to traverse a semantic graph populated with synthetic data from programming courses, students, and their grades, hosted in GraphDB. An open-source LLM, executed locally, transforms queries between languages and processes the query results to generate personalized and explanatory recommendations in natural language. The comparative evaluation measured the system's performance through two scenarios: execution times and accuracy. The results show that the system can offer personalized recommendations with 92% accuracy, which validates its potential to improve the learning experience through an accessible tool adaptable to educational contexts.