Research On Mining And Analysis Of Trajectories' Individual And Group Moving Patterns

With the rapid development of positioning technology,mobile terminals can collect users' location information more conveniently,thereby gradually forming massive spatiotemporal trajectory data.Mining the users' movement rules and trends from trajectory data becomes a hot spot of current research,which can be widely used in users' location prediction,traffic prediction,anomaly event detection,etc.Currently,individual users' semantic behavior pattern prediction and group moving pattern mining are two major research hotspots in the field of spatiotemporal trajectory mining.However,the former mostly uses over-sparse check-in data sets,while the latter is mostly centralized processing and is not suitable for real-time streaming environments.So we propose a location recommendation framework based on individual semantic behavior prediction,as well as a method to mine aggregation moving patterns in a distributed environment.The work is as follows.1.To predict individual users' semantic behavior pattern,we propose a new location recommendation framework based on predicting semantic behavior.Firstly,raw trajectories are transformed into semantic trajectories.Secondly,we cluster users and construct a Bidirectional Long-Short Term Memory(Bi LSTM)for each user's cluster to predict user's next semantic behavior.Thirdly,a Similarity-based Markov Model is constructed to obtain candidate recommended locations.Finally,a query operation is performed to find final recommended locations that satisfy both the predicted semantic behavior and the spatial distance constraints.This method overcomes the drawbacks of relying solely on over-sparse check-in data sets,and also makes full use of the spatiotemporal and semantic information of trajectory data.2.To mine the group moving patterns and analyze the evolution,we propose a framework for mining aggregation moving patterns and predicting evolution trends in a distributed environment.First,each distributed node constructs a two-layered distributed spatial index to facilitate the conversion of the movement snapshot of each time slice into the cluster snapshot.Based on the real-time cluster snapshot,the local aggregation moving patterns are incrementally mined through clusters' similarities.The coordinator node receives the local moving patterns sent from the distributed nodes and merges them into global moving patterns,and combines the users' next predicted locations to predict the evolution trends of moving patterns.The main contributions are as follows.On one hand,for the prediction of the semantic behavior pattern of individual users,we mine the potential semantic information of GPS trajectories,and propose a Similarity-based Markov Model,and the predicted semantic behavior is combined with SMM to recommend locations.On the other hand,for aggregation moving pattern mining and evolution analysis,we propose a two-layered distributed spatial index to improve the clustering process,and that the distributed nodes and the coordinator node collaborate to mine aggragation moving patterns and predict evolution trends.Theoretical analysis and experimental results demonstrate the supriority of the proposed framework and method.