Research On Semantic Frequent Pattern Mining Algorithm Based On Trajectory Data

Frequent pattern mining of trajectory data aims to mine a set of regions frequently accessed by users from a large amount of trajectory data,and it is the foundation of the application of a series of location-based services.Semantic trajectories have higher quality and less volume,It is better to reflect user behavior.They have attracted widespread attention.However,in the research of trajectory-based semantic frequent pattern mining,there are still problems such as inaccurate trajectory semantic labeling and incomplete mining patterns.Aiming at the current shortcomings of trajectory semantic frequent pattern mining,a research will be carried out.The main research contents of this thesis are as follow: 1.Aiming at the problem that the stay points detection accuracy of sampling irregular trajectories is not high,a stay points detection algorithm based on time series clustering is proposed.Firstly,based on the data field theory,a mixed characteristic density measurement method considering the spatiotemporal characteristics is designed.Then,based on the characteristic that the center density of the stay points is greater than the entrance,the filterrefine strategy is used to extract the stay points.In the filtering phase,points that are continuous in time and meet the minimum density threshold are used as candidate stay points.In the refining phase,the actual stopping points are screened by the maximum threshold.Experimental results show that the method can effectively detect the stay points in the irregular sampling trajectory,and have higher accuracy and lower time consumption than the existing methods.2.Aiming at the problem of fuzzy location of stop points,a location matching algorithm based on the hidden Markov model is proposed.The algorithm first marks the original trajectory record as a stop or move,and treats a continuous stop as a stop event.Then,using the real-world location repository,each stop event is associated with a potential candidate stop location.Since there may be multiple location candidates for each stop event,combining the improved Term Frequency-Inverse Document Frequency and Hidden Markov Model,each stop region sequence is matched with its most likely visited real-world location sequence.In addition,in order to solve the problem of low accuracy of traffic pattern recognition,a traffic pattern recognition method based on ensemble learning is used.Experimental results show that the proposed algorithm is superior to the existing improved algorithms.3.Aiming at the problem of incomplete mining of semantic trajectory patterns,a generalized sequential pattern mining algorithm and prefix projection pattern growth algorithm under multi-dimensional semantics are proposed for frequent pattern mining of multi-dimensional semantic trajectories.These two algorithms improve the frequent pattern mining algorithm based on GSP and Prefix Span,respectively,and design a support calculation method that takes into consideration the transition time interval and multi-dimensional semantic information.At the same time,the time interval is added when constructing the pattern.Experimental results show that these two algorithms can extract more frequent patterns and achieve better mining results.