Algorithm And Platform Research For Mobile Users' Behavior Analysis

In today's era of technology and network, the collection and preservation of human behavior data becomes easier, which creates favorable conditions for the study of human behavior. Call detail records, as a kind of human behavior data, truly record mobile users' daily activity trajectories, and can reflect human temporal-spatial behavior patterns, thus having great research value.In order to effectively and efficiently study mobile users' behaviors, a platform for mobile users' behavior analysis is designed and implemented in this paper. This platform can provide the whole function of dealing with mobile users' behavior data, including data management, data analysis, data calculation and the visualization of data mining results. In addition, the platform makes full use of the advantage of data visualization technology in knowledge discovery, and realizes an effective visualization of call detail records.With the aid of the platform, we further study call detail records. We focus on mobile users' activity trajectories and analyze the hidden characteristics of mobile users' behaviors. Two kinds of trajectory characteristics based user identification algorithms are then proposed. The key of trajectory characteristics based user identification algorithm is to find an effective characterization of mobile users' trajectory characteristics. The first approach introduces the location probability distribution vector to represent users' trajectory characteristics. This approach takes advantage of the spatial information of a trajectory. On the basis of the first approach, the second approach joins the temporal relationship between locations, and introduces the location shift frequency matrix to represent users' trajectory characteristics. With the characterization of trajectory characteristics, we can get the similarity between their corresponding trajectories by measuring the similarity between two trajectory characterizations. The similarity between two trajectory characterizations is calculated by cosine similarity in this paper. Finally, we argue that two trajectories with highest similarity are more likely to belong to the same user, so user identification result can be got accordingly.To verify the validity and applicability of the algorithms, they are tested under different conditions, including controlling the length of the trajectory through choosing different time slots and controlling the density of sampling points of the trajectory through choosing different mobile user groups of different call frequencies. The experimental results show that the algorithms have a higher identification accuracy under the condition of better quality of trajectory data, and certain validity is also guaranteed under the condition of poor quality of trajectory data.