HGDCN: A Multi-Scale Representation Learning Method for Knowledge Discovery and Social Networks

Mengjie Qian, Yunfei Chen, Jing Li

Abstract: The popularity of social networks is profoundly affecting people’s work and life. However, the social relations embedded in online social networks are extremely rich and have far exceeded the paradigm of ordinary binary relations, so the graph neural network models constructed based on binary relations have been difficult to deeply mine these implicit higher-order social relations. To obtain a more multi-scale online social network representation capability, this paper studies the multi-scale online social network representation based on hypergraph structure perception, convolutional neural networks, and KNN clustering. A Deep Clustering Network for Hypergraphs (HGDCN) is proposed for knowledge discovery and multi-scale representation learning in social networks. The hyperedges in our proposed algorithm can naturally express various environmental information in online social networks, which can more effectively simulate the interaction between users and the environment and improve the representation ability of online social network models. The experimental results show that our proposed HGDCN can achieve social network group cognitive features and combine information dissemination and fusion.

External IDs:dblp:journals/tce/QianCL25