Диссертации, представленные на защиту и подготовленные в НИУ ВШЭ

The thesis investigates graph embeddings for machine learning on structural data. It highlights critical privacy vulnerabilities in static graph neural networks (GNNs), showing that training links can be extracted through membership inference attacks. To mitigate these risks and better reflect the evolving nature of real-world systems, the research transitions to dynamic graphs. A unified benchmarking framework is introduced to enable fair evaluation of dynamic GNNs, addressing inconsistencies in prior studies. Based on this framework, the re-evaluation of leading models has uncovered new insights into their performance. Finally, the thesis develops two novel edge-centric architectures that directly model temporal dependencies through interaction patterns, achieving superior performance by capturing fine-grained dynamics often missed by node-centric approaches.

Nowadays, recommender systems are essential components of various consumer services, from e-commerce to social media. They help navigate through a large volume of items empowering user experience. Methods vary from classic matrix completion techniques to modern sequence models inspired by natural language processing. One of the prominent approaches is to consider the recommender system as a link prediction problem on a bipartite user-item interaction graph.The dissertation studies the adaptation of network embedding techniques to recommender systems. In the dissertation, we investigated different aspects of graph machine learning techniques, and their effect on downstream link prediction problem. We proposed an efficient strategy to incorporate node and edge features with structural information to solve the link prediction (recommendation) problem. Also, we developed novel models to preserve temporality and enable graph-based exploration for recommender systems. Proposed approaches were compared with the state-of-the-art open-source benchmarks and showed the efficiency of our approach.