Node Similarity Index and Community Identification in Bipartite Networks

Abstract

Bipartite networks or affiliation networks are a particular class of complex networks. It comprises two types of nodes, and only edges between the nodes of different types are allowed. The bipartite network model is a natural representation of the relationships between diverse entities. Most of the traditional complex network research focuses primarily on a single network, so research on bipartite networks is particularly necessary. In this paper, a novel DA similarity is proposed to measure the similarity between nodes, which takes both the influence of nodes and neighborhood structure information of nodes into consideration. Based on the DA similarity index, a community detection algorithm for bipartite networks (CDBNS), is firstly proposed. The experimental results show that DA similarity is superior to traditional similarity indices, and the CDBNS algorithm has an excellent performance in modularity and time-consuming. Furthermore, we employ the CDBNS algorithm in recommendation tasks and propose a recommendation algorithm called RASCS, which calculates the node similarity of each community detected by CDBNS and incorporates user-based collaborative filtering to achieve recommendation. It is also verified by experiments on several real-world datasets that the RASCS algorithm outperforms some baselines, such as RACD, ItemBasedCF, and UserBasedCF algorithms.