Abstract:

Graphs are a natural way to model complex information in many domains, including social networks, recommender systems, biological networks, communication networks, and the Web. Understanding graph structure and the processes that unfold on graphs, such as information diffusion, are fundamental problems that have attracted the attention of researchers across many fields.

In this talk, we will present algorithms for studying both graph dynamics and graph structure.

The first part will address network dynamics through the Influence Maximization with Link Recommendation problem. Building on the idea that graph structure affects how information spreads, we will present a link recommendation algorithm designed to improve information diffusion by suggesting a limited number of new connections among nodes.

The second part will focus on attributed multi-relational networks, where nodes model entities, edges capture multiple types of relationships, and nodes may be enriched with features, attributes, or labels. We will introduce an embedding approach that jointly represents nodes and their attributes in a low-dimensional space, allowing network topology and attribute information to be analyzed together. We will then show how this representation supports downstream tasks such as node clustering and semi-supervised node classification.

For both problems, we will discuss theoretical foundations and experimental results on real-world networks