User Activity Inference And Behavior Pattern Analysis Based On Mobile Phone Data

Recently,with the rapid development of smart mobile phone and Internet,the huge amount of location information of users can be obtained and stored.Users' activities in real world leave traces in space such as going out for eating,shopping,entertainment,which can reflect the lifestyle and interest preference of people.Among the location data,mobile phone data are the most persuasive data for characterizing people's life,because of its advantages of covering a wide range of people,long observation and cheap to get,more researches are devoted to study human behavior with mobile phone data.Through the study of user's trajectories,we can have a better understanding of user behavior and it's crucial to personalized service and advertisement.Due to the demand,this thesis designs and implement a user activity inference and behavior analysis system based on Spark distributed platform,which can automatically implement some functions such as data cleaning,trajectory extraction and collecting POI.After the preprocessing,we implement the function of user activity inference and based on this,we integrate the user behavior analysis model on this system,including the extraction of visited location number,radius of gyration,step length of each activity,temporal rhythm of activities,daily trajectory patterns and daily activity patterns.In the process of the realization of the activity inference,first,we look into the existing research and analyze the problems in current researches.Considering the data bias in POI and the different influence of POI on user activity under the different topic of region,this thesis introduces the topic generation model to extract the topic of POI of stay points in the city,then we model the correlation of POI and user activity with the POI distribution in the topic to infer human activities.After this,we explore the effect of multiple factors with survey data.Finally,our model is validated against the survey data and proved to be better compared to existing methods.Based on the result of activity inference,the thesis design a set of extraction algorithm and analysis method of user behavior.We divide user behavior into short-term and long-term behavior,at the same time,we consider both the mobility and activity dynamics of user to obtain a full analysis of user behavior.With our method,we can better understand user's lifestyle and interest preference and provide more personalized service for user.