Abstract

It is a common phenomenon in real life that individuals have diverse member relationships in different social clusters, which is called overlap in the science of network. Detecting overlapping components of the community structure in a network has extensive value in real-life applications. The mainstream algorithms for community detection generally focus on optimization of a global or local static metric. These algorithms are often not good when the community characteristics are diverse. In addition, there is a lot of randomness in the process of the algorithm. We proposed a algorithm combining local expansion and label propagation. In the stage of local expansion, the seed is determined by the node pair with the largest closeness, and the rule of expansion also depends on closeness. Local expansion is just to obtain the center of expected communities instead of final communities, and these immature communities leave only dense regions after pruning according to certain rules. Taking the dense regions as the source makes the label propagation reach stability rapidly in the early propagation so that the final communities are detected more accurately. The experiments in synthetic and real-world networks proved that our algorithm is more effective not only on the whole, but also at the level of the node. In addition, it is stable in the face of different network structures and can maintain high accuracy.

Highlights

  • We propose an overlapping community detection algorithm named

  • LELP that is a combination of local expansion and label propagation

  • The experiments in diverse synthetic networks and real-world networks show that LELP is excellent, compared with other algorithms

Read more

Summary

Introduction

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. There are extensive works on detecting overlapping communities [7] for which we provide a more detailed introduction . These studies are still insufficient in some ways. The static local optimization goal for detecting overlapping communities is adopted generally, such as fitness function in local expansion [8], optimization for modularity [9] and local Nash equation [10] in the game algorithm. The density, scale and edge distribution of different communities in real social networks are diverse. The immature communities detected by local expansion are pruned to be dense regions in order to conduct label propagation, a speaker–listener-based information propagation process. The experimental results show that our algorithm generally performs well and is stable in different conditions

Related Works
Algorithm
Evaluation Criteria
Synthetic Networks
Real-World Networks
Conclusions
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call