Similarity Measurement And Recommendation Method Research Based On Big Trajectory Data

With the development of global positioning technology,sensor networks and smart mobile terminal,a large number of trajectory data are accumulated.Trajectory data contains a wealth of information,including spatial,temporal,and other external descriptive attributes,such as text,pictures,video,audio,and so on.The trajectories including the locations,timestamps,and descriptive attributes are called multi-attributes trajectories.Trajectory analysis and mining show the great value,such as traffic route planning,urban planning,and commercial location.The major work includes:First,four features of big trajectory data,such as high dimensional heterogeneity,multi-granularity,uncertainty and high redundancy with noise,are summarized.According to different data types,the related research work of trajectory similarity measurement is summarized.The well-known measurements on spatial similarity,text similarity and time similarity are summarized w.r.t.the different data type.For trajectory recommendation,the related work of recommendation methods is summarized,such as the commonly used collaborative filtering recommendation method and model-based recommendation method.Second,two trajectory similarity measurements have been proposed.Most trajectory similarity work only focuses on the spatial-temporal features,and only a few work takes into account other types of information in the trajectory.For the trajectory similarity calculation,two new trajectory similarity measures,namely MMTD(maximum-minimum distance)and SUMTD(sum of minimum trajectory distance)are proposed and analyze the correlation among the spatial-temporal similarity and textual similarity with real dataset.The sensitivity of the similarity measure to the trajectory transformation is verified by simulation data.The measurement validity is verified and visualized through clustering by both a simulation dataset and a real dataset.Thirdly,existing trajectory recommendation methods usually consider certain cost criteria,such as shortest distance,minimum time,location popularity,etc.,ignoring user requirements.There are also some works that require users to accurately provide locations or requirement(textual)collections,increase user burden.To solve the problem,we propose two user behavior probability learning methods,one is based on matrix decomposition,and the other is based on multi-factor probability density function.The trajectory recommendation is made based on the probability of user behavior.Experiments were carried out using real trajectory data to verify the effectiveness of the algorithm.Finally,using the real Foursquare dataset for experiments and verification,the results show that the similarity measure algorithm can support approximate similarity for multi-attributes trajectories,and the trajectory recommendation method based on user behavior learning can improve the accuracy of recommendation.In future work,the proposed method can be applied to user group identification,friend recommendation or travel route recommendation system to improve user satisfaction and recommendation accuracy.