Heterogeneous Network Embedding With Enhanced Event Awareness Via Triplet Network

Abstract

Network analysis is an unavoidable topic in data mining today, and network embedding is an important means to help solve network analysis. With the increasing of network data volume, the content is increasingly complicated, the embedding scenario of homogeneous graph has been gradually replaced by heterogeneous graph. More and more embedding algorithms for heterogeneous graphs are proposed. Heterogeneous network can naturally integrate different aspects of information, so heterogeneous network embedding is a relatively effective method to solve the diversity of big data. It is helpful in the areas of anomaly detection, user clustering and intent recommendation. Here we propose a Siamese Neural Network optimization method based on event relations and meta graphs. This method ensures the semantic integrity and event integrity of heterogeneous graphs by using events and meta graphs respectively. Then put the graph information in Triplet Network for training, and the embedding results are produced. A classification task on a dataset for the true network are designed to prove the method. A real network data set classification task is designed to prove that this method is helpful for heterogeneous graph analysis.