User Behavior Analysis Based On Activity Trajectory Embedding

Due to the popularity of smart devices and wireless communication technologies,a large number of spatiotemporal trajectories describing the motion history of objeets have been generated.Among them,the activity trajectory enriches the traditional trajectory data by associating the activity information with the space-time information.These trajectory data contain a wealth of user activity information that deepens our understanding of human behavior.However,the existing research on space-time trajectory ignores rich user activity information and lacks personalized analysis of user behavior.Therefore,in this paper,we use the time and activity attributes of the activity trajectory to model user activity,analyze user behavior similarity,and predict user behavior status.Overall,the main contributions of this article are as follows:(1)In the user behavior similarity task,the past work mainly uses the time and space information of the activity place,ignoring the activity information that is most closely relat-ed to the user behavior.Therefore,this paper proposes a user behavior similarity algorithm based on activity trajectory embedding.The algorithm first uses the Gaussian mixture model to fit the activity time distribution and segment the time,then scans the user activity trajecto-ry according to the sliding window to obtain the user and trajectory features.A hash-based similarity ealculation method is used to eompare behavioral similarities between users.A large number of experiments have verified the performance and efficiency of our proposed method.(2)In the user behavior prediction task,the existing work mainly predicts through the user historical activity record and the social relationship,and lacks the mining of the user behavior preference.Therefore,this paper proposes a user behavior prediction algorithm based on active trajectory embedding.The algorithm first extracts user behavior prefer-ence features through paragraph vector model,then combines user behavior preferences and historical activity records,and analyzes each feature's impact on behavior state through at-tention mechanism to predict next behavioral state.Experiments on real datasets show the performance and efficiency of our approach.