Towards Trustworthy Graph Neural Networks and Their Applications in Recommender Systems

Graph Neural Networks have demonstrated remarkable success in modeling graph-structured data and are increasingly applied in various domains, including recommendation, drug discovery, and financial analysis. However, concerns regarding their trustworthiness, due to issues like data noise sensitivity, interpretability, and fairness, hinder widespread adoption. This research aims to enhance GNNs trustworthiness from three critical perspectives: reliability, explainability, and fairness. We introduce novel methods, including incorporating logical reasoning and model calibration techniques to ensure robust and reliable predictions, employing neural architecture to search for effective explanations, and analyzing the uncertainty quantification in fair GNNs with various strategies. This work establishes foundational insights and strategies for developing trustworthy GNNs, paving the way for equitable and transparent predictions in high-stakes applications.