A New Novel Label Propagation Algorithm

Abstract

Community detection is an enduring research hotspot in the field of complex networks. The label propagation algorithm is a semi-supervised learning method, which has the advantages of close to linear time complexity, simplicity and ease of implementation. However, LPA has two significant shortcomings in dividing communities: poor accuracy and strong randomness, which seriously affect the performance of the algorithm. This paper proposes a new label propagation algorithm to solve these two problems. In the initialization stage, a new node importance metric is proposed, which simultaneously considers the importance both of the node itself and its neighbor nodes to rank the importance of the nodes. In the label propagation stage, We also propose a new node similarity metric and the label is updated according to the similarity between the current node and neighbor nodes. Our experiments on real networks and artificial synthetic networks show that this algorithm can effectively find community structure and has better stability and accuracy than some existing LPA improved algorithms, and this advantage is more obvious on large networks.