Context Aware Check-in Prediction Based On LBSN

With the popularity of mobile intelligent terminals,mobile social networks have sprung up.People could obtain their position anytime and anywhere through diversified positioning technology.Location-based social networks(LBSN)have emerged.The users of LBSN generate many location data.It provides an opportunity to study the behavior patterns,daily routine and personal preferences of users and provide high-quality personalized service.This paper analyzes and predicts the behavior of LBSN users from the frequency,temporal,spatial pattern,social and weather factors.The related problems of user check-in behavior discussed and studied intensively.The results are as follows:1.Through analyzing frequency feature,temporal and weather feature,this paper proposes a location prediction model based on the weather and temporal feature.First,frequency feature modeled by markov chain mode,then combine with temporal preference to revise forecast results.Finally,fuse the location weather feature to obtain final prediction results.Experiments on two real check-ins datasets show that WTMC is 5% to 10% higher than baselines.2.This paper proposed user dissimilar rate to measure the degree of deviation of user check-in location.The analysis indicates different diverse rate user has obvious difference at temporal-spatial aspects,based on this discovery this paper presents a novel hybrid model,namely based on user diverse rate personalized check-in location prediction model(UDR).UDR model consists of two sub-models: First sub-model is User's Dissimilar Location Prediction Algorithm(UD).It considered temporal,spatial and social features,in addition considered the "hot-cold location transfer" feature,weather factor.This sub model predicts location where users have never been.Second sub-model is User's Recursion Location Prediction Algorithm(UR).Predict the probability that users return to the history check-in locations.The experimental results on two LBSN data sets verified that UDR was superior to the stat-of-the-art methods and the classic prediction models.