Community Discovering Using Social Links And Temp-spatial Topics In LBSN

In recent years, with the popularity of the Internet, online social networks develop rapidly. Due to its high amount of data and authenticity, new opportunities of many disciplines and techniques have been created. Community discovery is one such technique, which mainly studies the clustering reflected by complex relationships between users in social network. The traditional concept of community refers to the connection between the internal nodes are relatively close, but the connection between different communities is relatively sparse. However, with the abundance of contents in online social network, more and more factors need to be considered about relationships between users and discovering community. Especially, Due to the abundant information of off-line behavior, the emergence of location-based social network brings new challenges to community discovery.To address the problem, this paper studies multi-view community discovery in LBSN from the perspective of heterogeneous networks. Communities discovered in this paper must meet two requirments. The first requirment is frequent user interaction within the community, and the second is consistent user behavior pattern, which includes temporal pattern, spatial pattern and interests.Firstly, we propose a new measurement of dynamic social relationships based on the definition of user interaction in LBSN, and discuss temp-spatial topics from the perspective of temporal regularity, spatial regularity and interests. Secondly, we build a novel community discovery model to describe the process of network data based on social relationships and temp-spatial topics. Thirdly, we derivate its hidden varible sampling rules and multinomial parameter updating rules. Then, the Gibbs sampling algorithm for our model can be completed. Based on the model and algorithm, we design and implement the system of discovering community in LBSN.We select foursquare as the data source, and filter it to get checkins in New York as the dataset in our experiments. The results of community discovery are shown and analyzed from the perspective of user interaction, temproal regularity, spatial regularity and interests. So it is proved that our model can meet the two requirments well. In addition, we compare our model with TURCM, and the results demonstrate that the perplexity of our model is lower and our model fits the community structures in the dataset better.