Dynamic Community Evolution Analysis Framework for Large-Scale Complex Networks Based on Strong and Weak Events
Shaojie Qiao,Nan Han,Yunjun Gao,Rong-Hua Li,Jianbin Huang,Heli Sun,Xindong Wu
DOI: https://doi.org/10.1109/TSMC.2019.2960085
2021-01-01
IEEE Transactions on Systems, Man, and Cybernetics: Systems
Abstract:Community evolution remains a heavily researched and challenging area in the analysis of dynamic complex network structures. Currently, the primary limitation of traditional event-based approaches for community evolution analysis is the lack of strict constraint conditions for distinguishing evolutionary events, which entails that as the cardinality of discovered events increases, so does the number of redundant events. Another limitation of existing approaches is the lack of consideration for weak events. Weak events can be generated by small changes in communities, which are empirically prevalent, and are typically not captured by traditional events. To manage these two aforementioned limitations, this research aims to formalize a weak and strong events-based framework, which includes the following newly discovered events: "weak shrink," "weak expand," "weak merge," and "weak splity" predicated on the community overlapping degree and community degree membership, this article refines these traditional strong events, as well as new constraints for weak events. In addition, a community evolution mining framework, which is based on both strong and weak events, is proposed and denoted by a weak-event-based community evolution method (WECEM). The framework can be summarized by the following: 1) communities in complex networks with adjacent time-stamps are compared to determine the community overlapping degree and community membership degree; 2) the values of the community overlapping degree and membership degree meet the definition of events; and 3) weak events are effectively identified. Extensive experimental results, on real and synthetic data sets consisting of dynamic complex networks and online social networks, demonstrate that WECEM is able to identify weak events more effectively than traditional frameworks. Specifically, WECEM outperforms traditional frameworks by 22.9% in the number of discovered strong events. The detection accuracy of evolutionary events is approximately 12.2% higher than that of traditional event-based frameworks. It is also worth noting that, as the cardinality of the data grows, the proposed framework, when compared with traditional frameworks, can more effectively, and efficiently, mine large-scale complex networks.