Research On User Segmentation Modeling Based On Mobile Positioning Data

With the rapid development of mobile Internet,people are increasingly using mobile devices to access various kinds of services on the web.These devices record a large amount of spatio-temporal data of users.User movements are generally driven by either daily activities or interests.Staying at home or at a workplace can be determined as routine activities which a user follows in sequence.In addition,other visited places are mainly influenced by user's interests.If a user has high access frequency to a certain geographic location,it indicates that he has a higher degree of interest in this location.Therefore,mining these data can reveal user's behavior habits and interest preferences.In this paper,through systematic combing and analysis of existing research results at home and abroad,a user segmentation model based on mobile positioning data is proposed.Users with similar interest preferences are categorized into some group.Based on the results of user segmentation,users can be provided with customized location-based services including location recommendations,friend recommendations,identifying hidden social links,community detection,etc.First of all,we summarize four characteristics of the mobile positioning data recorded by the mobile device when the user is surfing the Internet.Secondly,according to the spatial and temporal heterogeneity,the spatial aggregation effect and the sparseness of these data,we propose the concept of Region of Interest(ROI),and use the Density-Based Spatial Clustering of Applications with Noise algorithm to extract the user's ROI.Thirdly,we use the AMap API to get all the Point of Interest(POI)fallen in the ROI,and employ the idea of TF-IDF to build the feature vector of each ROI according to the characteristics of the data whose semantic information is unknown.Then,we compute the weighted average of the feature vectors by taking the user's access frequency to each ROI as a weight and then construct the interest vector for each user.Finally,we use the cosine similarity as the measure of the similarity between users,and use the agglomerative hierarchical clustering algorithm to cluster the users into several groups.We perform experiments on a real-world dataset collected by 10 thousand users in a period of one month and compare the two clustering results obtained by the agglomerative hierarchical clustering and K-Means.Experimental results show that although there are some deviations in the clustering results obtained by the two clustering algorithms,the overall results are consistent.The probability that the same user is classified into the same group by two clustering algorithms is more than 85%.Therefore,it further verifies the feasibility and effectiveness of the proposed user segmentation framework in this paper.