Optimization Model And Efficient Algorithm For Temporal Community Detection

Data with networks structure is now a hot research in data science.Compared with static networks,the dimension of temporal data is higher and the structure of temporal networks changes flexibly.It is more difficult to establish mathematical models and numerical algorithms.This paper focuses on the community detection in temporal networks.Use low-rank optimization theory to get a minimum convex optimization model for temporal community detection.Based on the application requirements of community real-time detection,a data-driven temporal label walk(TLW)algorithm is proposed.Specifically,the contributions of this thesis mainly include:Firstly,the convex optimization model of static community detection is extended to temporal community detection.Result based on artificial data sets shows that our model can recover the community structure which static community detection methods and temporal community detection methods based on temporal smoothness can not recover.Secondly,a heuristic community detection algorithm TLW based on temporal walk and label propagation is proposed.Experiments on real data sets S tudents and Facebook show that compared with other data-driven community detection algorithms,TLW can get higher modularity with less time cost when the time window increases,so it has obvious advantages in computational complexity and community quality.The work of this paper is valuable for the analysis of command and control relationship of military data,information mining of networks data,commodity recommendation of online trading system,etc.