raph Neural Networks, from Theory to Recommender Systems

Dr. Walter anelli

Politecnio di Bari, Italia

Resumen/Abstract

We present an exploration of Graph Neural Networks (GNNs), bridging foundational theory with critical applications in modern recommender systems. We address a crucial challenge in the field: moving beyond the black-box application of GNNs to understand the empirical and theoretical reasons for their success. We introduce the core principles of GNNs, from message-passing paradigms to spectral graph convolutions, establishing the theoretical groundwork for this novel learning approach. Leveraging insights from recent reproducibility studies, we introduce a novel evaluation perspective that directly links model performance to the intrinsic topological properties of the user-item graph to interpret why certain models excel or fail based on dataset characteristics.  We conclude by showing the practical implementation of key graph collaborative filtering methods within the Elliot framework, a robust and reproducible toolkit for research and development

Profesor Proponente EPS: Alejandro Bellogín

Curriculum ponente

Dr. Vito Walter Anelli is a Tenure-Track Assistant Professor (RTT) at the Department of Electrical and Information Engineering, Politecnico di Bari. He earned his Ph.D. with a Doctor Europaeus certification and cum laude honors in Electrical and Information Engineering from the same institution in February 2020. His doctoral dissertation, "Knowledge-Enabled Recommender Systems in the Linked Data Era," received the SWSA Distinguished Dissertation Award from the Semantic Web Science Association in 2021. He was awarded the Best Student Research Paper at ISWC (CORE A) in 2019, the Runner-Up Short Paper at CIKM (CORE A) in 2021, the Best Student Research Paper at UMAP (CORE B) in 2025, and the Best Short Paper at SIGIR (CORE A*) in 2025. Dr. Anelli's research is situated at the intersection of Artificial Intelligence and Personalized Information Access. His primary interests encompass Recommender Systems, Machine Learning, Large Language Models in Information Access, Semantic Web and Knowledge Representation, and Federated Learning and Privacy in AI.