Tensor Based User Trajectory Data Mining

With the development of GPS-Embedded technology, Satellite Technology and Wireless communication technology, and the popularity of mobile devices, more and more location based services begin to enter people's daily life. In recent years, the proliferation of these devices has given rise to an increase in location-based services, people can upload and share their location and GPS trajectory through a variety of platforms and applications at any time. A large amount of GPS trajectories are accumulated in daily life and services for different types of applications.While people enjoying the convenience and benefits provided by these trajectories associated emerging technologies and services, this also bring a great challenge to the management and utilization of mass trajectory data. And most of applications still use raw GPS data like coordinates and timestamps, without much understanding. How to handle massive amounts of trajectory data efficiently, how to handle the sparsity of trajectory data efficiently, how to extract useful information from trajectory data efficiently and intelligently, mining the user and application related context information.Because of such problems, based on the ternary relation: source-destination-corresponding road segments, present a tensor based user trajectory data mining system. Using tensor to store, process and compute trajectory data, provide collaboration for different users' trajectory data, improve the integrity of the trajectory data. People can analyze and mine trajectory data from different view, and if use source and destination pair as input, system can find the hot route, this hot route can be used for user trajectory recommendation, user trajectory recovery, user trajectory prediction and so on.By conducting extensive experiments on a real trajectory dataset, the results demonstrate the solution can recommend a Collaborative hot route with considerable accuracy. The average hit rate is 59.73%.