Research On User Location Prediction Based On Sparse Spatiotemporal Data

With the gradual improvement of WiFi coverage and the development of location technology,a large amount of spatio-temporal location data based on Wi Fi logs can be obtained,which contains a wealth of information about human movement and promotes the rise of location prediction research based on historical trajectories.In real application scenarios,trajectory data usually presents the characteristics of sparse distribution and single structure due to low sampling,which leads to low accuracy in location prediction.In response to this problem,the paper has carried out the following research work:Firstly,a clustering algorithm for similar users that combines user attributes and trajectory information is proposed.Human attributes are related to personal characteristics and preferences closely,age,gender,education status,for instance,while historical trajectories reflect personal life,work,and rest.Therefore,the user attribute similarity and historical trajectory similarity are measured separately,and then the two are weighted and fused as the overall similarity of people.Based on this,all user groups are similarly clustered.The algorithm considers human attributes and historical movement information,which play an important role in alleviating data sparseness in location prediction and cold start problems.Secondly,a location prediction model SG-LSTM(Similar Group based LSTM model)based on similar user clusters is proposed.The low sampling of trajectory data makes it difficult to obtain the user’s complete trajectory route.Therefore,the sparse trajectory data is usually faced in the research of location prediction,which causes the model to fail to obtain sufficient historical information and the prediction accuracy rate is low.This paper divides people into similar clusters based on multi-source information and trains predict model with all trajectories in similar clusters,which can strengthen people’s historical trajectory information and provide a new method for trajectory prediction in sparse spatiotemporal data scenarios.Experiments on real Wi Fi spatiotemporal data sets show that the prediction accuracy of the SG-LSTM model reaches 88.9%,which is better than traditional models in terms of accuracy and training time.Thirdly,a position prediction model Seman Predict based on semantic information is proposed.Due to the improvement of positioning technology,it is convenient to obtain accurate location geotag information,the type of stay point activity and staying time.The semantic information above deeply reflects human activity patterns.Traditional prediction models mostly focus on people’s location transformations and changes in latitude and longitude,lack of attention to semantic information.In this paper,semantic information is using to train location prediction model.Experiments were performed on the real Wi Fi data set and Twitter social media data set respectively,and the prediction accuracy rates were77.34% and 67.14%,which are better than traditional trajectory prediction models.