Community Discovery Using User Social Relationships And Temp-spatial Behaviors In LBSN

In recent years,with the emergence of various social networks,new types of social services have been provided.Community discovery is a key technology,which aims to find user groups with similar features in social networks.The traditional community discovery algorithms often use topology connection information between users to find communities which require the nodes in the same community are closely connected,and nodes in different communities are sparsely connected.However,with the diversity of social data,the connections between users become more complex.Especially,the emergence of location based social network leads to the inapplicability of traditional community discovery algorithms and brings new challenges to community discovery because it integrates the location information into the traditional social network topology and makes it to be a heterogeneous network.Aiming at this problem,this thesis studies the multi-dimensional community discovery technology in LBSN from the perspective of reconstructing heterogeneous network.Community found in this thesis should satisfy the following three important properties: the users in same community are closely connected,and the geographic distance of their access areas are close,and their behavior patterns are consistent,which includes temporal pattern and interests.Therefore,this thesis proposes an LBSN isomorphic network model(LSHNM)which is based on the user social relations and temp-spatial behaviors to calculate the user similarity relation in multi-dimensional feature and construct LBSN isomorphism network topology.Then,non-negative matrix decomposition(NMF)is used to excavate LBSN communities which meet the above properties from isomorphism network topology,and finally the system of discovering community in LBSN is designed and implemented.This thesis use Foursquare New York data as data source.The experimental results of the LSHNM model and the comparison algorithms are compared from the aspects of social relations,temp-spatial distribution,behavior pattern closeness and operating time.Experiments show that the LSHNM model is more efficient and can find better community structure.