Representation Learning in Continuous-Time Dynamic Signed Networks
This work addresses the challenge of predicting polarization and signed structures in evolving networks, such as social or political interactions, by introducing a novel method for a previously unmodeled problem, though it is incremental in combining existing techniques for signed and dynamic networks.
The paper tackles the problem of modeling dynamic signed networks, which combine conflicting relationships and temporal interactions, by proposing a new GNN-based approach called SEMBA that incorporates signs using balance theory and evolves embeddings from higher-order neighborhoods. The result is that SEMBA significantly outperforms baselines by up to 80% on predicting future link signs and matches state-of-the-art on link existence prediction, with improvements attributed to better performance on the minority negative class.
Signed networks allow us to model conflicting relationships and interactions, such as friend/enemy and support/oppose. These signed interactions happen in real-time. Modeling such dynamics of signed networks is crucial to understanding the evolution of polarization in the network and enabling effective prediction of the signed structure (i.e., link signs and signed weights) in the future. However, existing works have modeled either (static) signed networks or dynamic (unsigned) networks but not dynamic signed networks. Since both sign and dynamics inform the graph structure in different ways, it is non-trivial to model how to combine the two features. In this work, we propose a new Graph Neural Network (GNN)-based approach to model dynamic signed networks, named SEMBA: Signed link's Evolution using Memory modules and Balanced Aggregation. Here, the idea is to incorporate the signs of temporal interactions using separate modules guided by balance theory and to evolve the embeddings from a higher-order neighborhood. Experiments on 4 real-world datasets and 4 different tasks demonstrate that SEMBA consistently and significantly outperforms the baselines by up to $80\%$ on the tasks of predicting signs of future links while matching the state-of-the-art performance on predicting the existence of these links in the future. We find that this improvement is due specifically to the superior performance of SEMBA on the minority negative class.