Social Circle Inference Based On Mobile Trajectory

With the rapid development of cellular networks and wireless communication technology,a variety of mobile devices and location-based social networks(LBSN)services have been widely used in our daily life,which generate huge amount of location related data,such as photos,tweets and check-ins.These data provide an unprecedented opportunity for us to discover knowledge from human mobility data that is an important task for many downstream applications such as point-of-interest recommendation,motion trace identification,personalized trip planning,etc.A specific problem that has recently spurred research interest is the so-called Social Circle Inference from Mobility data(SCIM),aiming at inferring relationships among users based on mobility data without any explicit structured network information.Existing methods either require partial social ties or fail to model the implicit correlations between user links,thereby suffering from critical inference bias.In this research,a novel SCIM framework,called SCIM via self-Attention(SCIMA),is presented and analyzed,which contains systematic human mobility mining methodology and social relation inference approaches.Instead of directly applying the recurrent model on training user trajectories as previous work,SCIMA encodes the semantic information of trajectories through self-Attention mechanism,and then predicts the corresponding label set of current trajectories.In particular,we find that the traditional word embedding technology can only learn a fixed embedded representation vector for each check-in point.In order to capture the context information associated with user check-ins in different trajectories,this paper introduces a new module for context-aware check-in representation learning by adaptively incorporating the internal states of the recurrent layers,which is more effective than the context-independent check-in embedding used in existing social circle inference frameworks.Towards that,an improved model SCIMAC--SCIM via self-Attention and Contextualized embedding--is proposed,which allows us to better identify the semantically similar motion patterns while effectively alleviating the circle bias inference problem.To model the underlying correlations between labels,SCIMAC further leverages a more sophisticated label embedding technique to adjust the penalties for correlated users,enabling a better understanding of the user interactions in the label space.We conduct extensive experiments on four real-world LBSN datasets to evaluate our framework from different aspects.The experimental results demonstrate that our proposed SCIMA and SCIMAC both obtain significant performance improvement on all datasets compared to the state-of-the-art baselines.In contrast,the performance of SCIMAC is improved more obviously because of the addition of context-aware check-in representation learning and the operation to reduce inference error on the basis of SCIMA.In general,social circle inference based on mobile trajectory is crucial to both users and businesses.Therefore,this research sheds light on both interpretable human mobility mining and LBSN-related business applications.