An Improved Label Propagation Algorithm for Community Detection Fusing Temporal Attributes

Abstract

Community detection algorithms are used to analyze the community structure. However, existing algorithms are based on undirected or undimensional network structure, which cannot make full use of the user’s attributes. Based on the academic social network, we propose an algorithm that integrates multidimensional network structures and user attributes, which can divide community by using an improved Label Propagation Algorithm (LPA). We construct a real social network by integrating multidimensional network, and use the similarity of research fields of scholars to generate edges between nodes. The directed edges are weighted by the four-dimensional network and temporal attributes, while the nodes are weighted by the social status and influences in the network structure of users. Then, the propagation probability of label is defined by the weighted indicators, and the community is divided via the directed weighted academic network. The experimental results show that the proposed algorithm LPA-STN can effectively improve not only the accuracy of the community detection results, but also the compactness within the communities.