User Movement Prediction Based On Social Relationship

With the wide use of positioning device,mobile phone and Location Based Social NetWork,more and more user location data is available.Mining the principle of user movement is of great importance in traffic control,road planning,resource allocation and so on.User movement is partly effected by social relationship.Considering the effect of social relationship when predicting user movement is important.The disadvantage of current neural network mobility predicting model is the lacking of long term memory,within the hidden memory of model parameters.It can't compare two users' trajectory and can't ming the relationship between users.Thus current neural network mobility predicting model can't mining the social influence on user movement.To cover the shortage and mining social influence,this paper gives FRMM(Friendship recurrent memory model)with the long term memory part to record users' trajectory,solving the shortage of current model.With the writing rule and reading rule to solve content writing and reading.With the addressing mechanism to compare current user's trajectory with it's friends' trajectory,extracting useful information.With the friendship impact factor to enlarge important friends' influence.This paper implement the method to predict user movement with friendship.It composites of three parts,data preprocessing,model implementation,model training and test.Data preprocessing extract numerical features from raw dataset,mining hidden friendship and split check-in data into trajectories.Model implementation part implement FRMM,give details of each part of FRMM.Model training and test part input numerical features and train model with forward and backward propagation to get generalized model and evaluate it.Choosing main efficient mobility predicting models as compared model,this paper gives experiments on three real-world user check-in dataset.The results show that FRMM outperforms the state-of-the-art neural network mobility predicting models.