Locations Forecast And Activities Recommend Base On GPS Trajectory

Through the GPS device, people can take full advantage of their own current geographical position as the context and the corresponding location-based services (LBS) to achieve the interaction. With the popularization of GPS-enabled devices, people get a lot of GPS track information; however, a great deal of users’information which is involved in GPS track information has not been effectively exploited. In recent years people gradually begin to realize this, so they conduct extensive track-related researches. But numerous researches are still limited to coarse-grained researches which are related to major points of interest.This paper put forward a location prediction algorithm based on the Markov model-a recommendation algorithm based on users’similar trajectory. The algorithm is divided into two main parts. The first step is to get a top-k similar users by a user-location Interaction Matrix, and then to query each user’s state transition matrix, ultimately to get the top-k possible locations. The main difficulty of the algorithm lies in how to build the user location interaction matrix and transition matrix. For the user-location interaction matrix, the paper proposes a partition-based user-location interaction matrix, namely through the stop zone (SR) clustering again to reduce the dimension of the matrix. For the user-location transition matrix, the paper builds a SR topology diagram based on in advance position history in the SR, and then extracts the location transition matrix from the SR topology diagram. This algorithm adopts the division the user-location interaction matrix, on the basis of the matrix it extracts users’eigenvectors to calculate the similarity. So this algorithm not only greatly reduces the computational burden, but also gets higher quality similar users.Despite the recommendation algorithm based on users’similarity in the trajectory has a better recommendation quality, but it does not dig deep between the tracks associated with the activities. In view of this, the paper proposes location prediction based on semantic rules and activities recommendation algorithm. To make the GPS track information enable to contain certain semantic information, firstly, this paper makes advantage of improved GPS define to expand SP semantic in order to get the SPA model, then based on the SPA models it takes advantage Apriori algorithm to shape semantic rules among activities. Through location prediction based on semantic rules and activities recommendation algorithm, and the location users put in, via the SPAG we can guess the user’s current location and the possible activities. And then via the SPAG we can choose the most likely semantic rule to predict the most possible location where the user may engage in the activities. Since the algorithm deeply digs semantic rules among activities, so it owns the higher recommendation quality. The experiments show that the accuracy of the historical position recommendation raises up by 30%, and the accuracy of the new position recommendation raises up by 10.7%.