Research On User Feature Extraction And Behavior Pattern In Location-based Social Network

Advances in Mobile Internet and Global Positioning System(GPS)technology have led to the rapid development of location-based social networks(LBSNs),thus massive check-in data has been accumulated.The LBSNs data implies a large amout of user activity features and behavior patterns,so the feature extraction of LBSNs data has become a hot research direction.Which is important to discover the value of LBSNs data for travel plan and urban development for further improving the location-based services.Based on the user's check-in records in LBSNs,this thesis studies the location and semantic features of users' check-in,and explores the user's urban behavior patterns.The specific work extends in the following three aspects:(1)Users' check-in based location feature extraction.The check-in data contains a large amount of location information.This thesis clusters the check-in points in the LBSNs to discover the POI(Point of Interest)as the access hotspot(the hotspot area where the user has a high access frequency),thereby extracting the location characteristics of the user.Considering the large data volume and uneven density distribution of LBSNs,two clustering algorithms are proposed in this thesis: 1)integrated clustering algorithm based on Meanshift and K-means;2)integrated clustering algorithm based on DPC and KNN.Compared with these classical algorithms,this thesis finds that DPC+KNN algorithm has the best clustering precision and can adapt to any shape cluster,while Meanshift+K-means has better adaption to the work of extracting POI from LBSNs data.(2)Text content based semantic feature mining.In order to discover the semantic feature of the user's check-in(understand the user's travel purpose),this thesis annotates the POI based on the POI location,semantic annotations are added to each POI to intuitively understand the user's behavioral semantics.Thereby this thesis proposes a text-based POI semantic annotation method.The main idea is to extract semantic rules of different categories from the text information in LBSNs,and to perform multi-category semantic annotation on POI through text matching.Compared with the existing methods,the annotation method of this thesis has certain advantages in the accuracy of labeling and algorithm efficiency of some categories.(3)User behavior pattern based identification of urban functional region.In order to explore the behavior patterns of users in the LBSNs,and to explore the application value of user behavior patterns,this paper proposes a method based on user behavior patterns to identify urban functional areas based on POI extraction and annotation.This method introduces the idea of probability topic model,and compares the function of the urban region to the "topic" of the article.By comparing the recognition results with the target city regional planning data,the author finds that the urban region functions extracted by the method are basically in line with the urban regional planning.Compared with other existing methods,the accuracy of the result in this thesis has a certain improvement.