Research On Entity Clustering By Multi-views Community Detection

In recent year,the form of social networking platform has been even more violent.Users can take full advantage of those social network sites and content sharing platforms to build their own social relations.Such,real-time communicating with others,publishing many articles and sharing resources with others.Therefore,it is necessary to distinguish several dense subgraph form social networks,also namely community detection.Community detection is researching hotspot in data mining,machine learning,graph theory and other fields.Today,the technology of community detection can be used in crime detection,protein function prediction,Web community discovery,document clustering,and so on.in the huge relationships of global social network,Some single view algorithms are imperfect.Because these algorithms' conditions of partitioning is lacking and the factors is incomplete.The other side,the traditional multi-relationals community detection method always select the same nodes that cause these instances can't reflect the relationship of real world and That means the applicability is very low.According to these problem above,this paper can proposes a new algorithm that can improve the traditional algorithm in various angles.In this paper,the main contribution of the points are as follows:(1)We will proposed a Two-stage Multiple Views Network Detecting Model.It different from the traditional model that can use mutual promotion in various perspectives of relationship to enhance accuracy and also use the cluster fusion to find the better global community.(2)We will proposed a co-MLSC algorithm that can use local co-training method to solve the no sufficient problem.Through this way we can enhance the accurate of clustering.(3)We will proposed a MGCF algrithm that can combine lable matching and voting method to solve the limit problem that each view must have the same nodes and same clusters.