Interpretable recommendation based on graph neural networks

Abstract

This project focuses on enhancing the explainability of Graph Neural Network (GNN)-based recommender systems by integrating Large Language Models (LLMs) and Explainable User Interface (XUI) design principles to deliver user-friendly, interpretable recommendations. Through the development of a LightGCN model paired with an three different LLMs, namely OpenAI’s GPT 4o-mini, Meta’s Llama 2.5 and Gemini 1.5, the system translates complex model predictions into natural language explanations, improving accessibility for novice users. By bridging technical GNN outputs with a user-centred design, this research addresses critical gaps in transparency and usability in XAI, demonstrating a practical approach for deploying interpretable, AI-driven recommendations in real-world applications.