Research On Movement-based Significant Location Mining And Application

With the increasing popularity of smart mobile devices, people's daily behavior can easily be collected and stored by location-record-enabled devices. The recorded data not only identify users' physical positions but also contains a great of knowledge. If mining and analysis the significant locations from the physical locations, users' interests and preferences, users' work, distribution of city resources and so on can be extracted. Therefore, research on the implied information of significant locations has theoretical and practical meaning.The previous significant location mining algorithms based on clustering, filtering or mathematical model methods have several drawbacks such as low accuracy, complex computation, weak error tolerant and low interpretability. Therefore, this paper puts forward a significant location mining algorithm based on the movement characteristics, called MMC(Mining based on Movement Characteristic). First of all, according to user's move features in real world, the critical threshold of the “wandering” state is determined; secondly, the instantaneous speed of users' each position is calculated in the track, and the continuous points whose speed is higher than the threshold of the “wandering” state are excluded, then candidate locations by each wandering trajectory are generated; finally, the same locations are merged for the significant location list. In order to verify the effectiveness and correctness of the algorithm, this paper used the GPS trajectory dataset in the real world collected by Geo Life project. First, users' significant locations are mined in the dataset by MMC, compared with some other representative algorithms, the experimental results show that, MMC not only can be more accurate in mining users' significant locations from the GPS trajectory, but also has a higher efficiency.Based on the results of significant locations mining, the application of MMC algorithm, find similar users according users' significant locations list, in recommendation system has been researched. First, taking user's locations as features and taking account of location-visit frequency and the visiting time interval, the user-location correlation is defined to measure the relation between the user and the significant location; Secondly, based on all the significant locations in the list of users,users' relevant vectors are constructed to describe user's characteristics; Thirdly, each user's relevance vectors are calculated based on the union set of their significant location list; Finally, according to the product rule of user-location correlation and the relevant vector, users' similarity are calculated. Then we use the algorithm to finding users' similarities based on MMC's results, comparing with the distribution of users' locations on the map, the results show that algorithm perform better in finding similar users.