Using Graph Neural Networks for Heterogeneous Event Classification

Graph Neural Networks (GNNs) are specialized machine learning models designed for analyzing and making predictions based on graph-structured data. In this paper, we introduce a GNN-based approach for the analysis of heterogeneous event graphs. By heterogeneous event graph, we refer to a temporal graph where multiple types of nodes representing events as well as multiple type of edges among events are defined. A distinguishing aspect of the proposed approach lies in the specification of the subgraph of nodes for each event to classify, and in the use of embedding techniques for representing all the relevant features/attributes of the subgraph. The proposed approach is compatible with different GNN models; the use of PSCN, namely a CNN for subgraph classification, is discussed in the paper under two different settings. Preliminary results on two different case-studies, as well as a real application in the field of historical data classification are also discussed.