An Attentional Recurrent Neural Network Model Integrating Social Influence For User Mobility Prediction

User mobility prediction is an important application of location-based service networks(LBSN).In the current,traditional ways are no longer suitable for large datasets,allowing for some reasons such as the variety and complexity of the users' data.Methods based on neural network models are also facing some problems such as poor accuracy.There have been some researches proven social influence on user mobility.By introducing social influence into neural network modes in a proper way,the model can capture the similarity between users and their friends,which is of great significance.By integrating social embedding,mobility regularity extraction and location prediction,the Social Attentional LSTM Model(SALM)is designed.To solve the complexity of users' data,a recurrent neural network is used to capture the complex regularity.The attention mechanism is also used to highlight the locations that are related to the current location.To solve the huge difference among users' data,the user information in each check-in is embedded into a user representation vector,which is unique to every user.The vector can lead the model to learn the mobility regularity of different users respectively.To deal with the sparsity of users' data,social intimacy is defined to obtain users' friends.When generating user representation vectors,social influence is considered,which can help the model to capture the similarity between users and their friends,thus improving the accuracy of prediction and mitigating the impact of sparsity.Experiments have been conducted on several real datasets.The results have shown that SALM achieves higher accuracy comparing to the recent models,and social influence plays a positive role in SALM.