User Mobility Prediction Algorithm And Application Based On Crowd Sensing

With the development of communication technology and the popular of location devices,massive user trajectory data are generated.Through deploying historical data analysis,user modeling,and mining the relativity between users' mobility features and location,the spatial and temporal pattern in user mobility trajectories can be discovered.In this context,mobility prediction research draw more and more attention in recent years.In crowd sensing computing,mobility prediction can help better perceive users' spatial and temporal information,and further provide better basis for upper layer application design and deployment.At the same time,social interaction behavior is the primary component in human social activity,and it is an important way to facilitate social information sharing and social collaboration.So,in recent years,diversity applications and systems are developed to facilitate people's social interaction,promote users' cooperation and reinforce social relationships.This paper concentrate on mobility prediction and social interaction research in crowd sensing,and then propose a accidental social interaction supporting framework based on mobility prediction,in which accidental interactions means occasionally occurred,unexpected,have no affect to users' early established schedule,spatial and temporal intersection on users' trajectories.Mobility prediction models are generated through mining GPS trajectories data,and accidental interaction opportunities are figured out by participate in interaction activities using specific application in physical world.Generating new GPS trajectories through massive raw GPS data preprocessing,such as outliers deletion,Place extraction and Place clustering.Then,training dataset are composed from these new trajectories data after feature extraction process,and benefit from the spatial and temporal regularities existed in people's GPS trajectories data.This paper adopt supervised learning algorithms,such as decision tree,to train prediction models from empirical mobility patterns.After a comprehensive model evaluation and feature selection process,the finalized prediction model utilize users'current mobility context(e.g.,time,location)as inputs,to forecast user's future next venue,arrival time and user encounter information.Compared with probability statistics approach,the prediction model proposed in this paper reach more accuracy.Based on multiple users' prediction results,accidental social interaction opportunities,which occurred unexpected and have no effect on users' schedule,can be perceived.Utilizing these predicted interaction opportunities and according to the specific application scenarios in campus,this paper designed and developed a prototype application Buy4Me to facilitate serendipitous social interaction to encourage users to participate in actual activities in physical world,which is a mutual application that a user requests others to buy what she/he needs incidentally on their way.