Research On User Behavior Trajectory Clustering And Its Application

With the rapid development of the national economy and the acceleration of urbanization,traffic congestion has become a global problem that affects the sustainable development of the city.Traffic information service system as an important part of intelligent transportation can provide fast and effective traffic flow information,which can provide convenience for public-travel,alleviate the problem of traffic jam,improve road capacity,reduce the number of traffic accidents,cut down energy consumption,lighten environmental pollution and meet the needs of the urban harmonious and sustainable development.Data shows that current traffic information service system has multiple channels,such as radio,variable information board,web sites and mobile phone text messages,and the contents of traffic information are more rich and accurate.But for traffic managers and travelers,the current traffic information service level has not reached the demand of traffic participants.In order to improve the efficiency of people's travel further and alleviate the problem of traffic jam,academia and industry recently proposed an idea that build the traffic information service platform based on the smart phone,which can accurately describe public travel behavior characteristics through analysis of the acquired data(such as the historical data of mobile phone users check-in),in order to provide users with the appropriate travel routes.One of its key technologies is to design a suitable clustering algorithm based on user behavior trajectory.The main research contents of this paper include the following points:1.This paper proposes a user behavior trajectory clustering method based on check-in data.Compared with K-MEANS clustering algorithm,the proposing method considers the time dimension,and expends the comparison of the point object similarity measure in K-MEANS clustering algorithm to the linear object(user behavior trajectory).At the same time,when define the cosine similarity between users,put the check-in time,check-in date into the traditional matrix "users-check-in location",which became the "user-check-in time(date)-check-in location " cube.In addition,when updating clustering center,the new cluster center will be selected by the user who has the largest sum of similarity.2.In order to reflect differences in the number of the user check-in in different location and the evolution trend of user behavior trajectory,our algorithm fully considered the influence of the marginal effect of check-in date and the differences of check-in number during definition of the users' check-in value.If the same location has more times of user check-in,the location has the more important level in user behavior trajectory.At the same time,users' spatial-temporal behavior will be changed over time.If the check-in date is more close to the current date,which can reflect the user's current behavior.3.In order to achieve the parallel processing of large data,this paper proposes a parallel model of user behavior trajectory clustering method based on check-in data.At the same time,the optimal clustering number,algorithm convergence rate and other experiments have been analyzed by using real check-in data set,and clustering results have been evaluated by using the clustering evaluation indexes.