User Behavior Prediction And Service Recommendation Based On Location Applications

With the widespread use of location-based social networks(LBSNs),users can share their points of interest(POIs)by checking in.By analyzing users’historical check-in records,POIs recommendation can recommend places that users may be interested in so that they can get a better access experience.In LBSNs,because the proximity of geographical location has a significant impact on the check-in behavior of users,social friends often have common interests,so geographical influence and friend influence are widely used in location recommendation.In addition,people also prefer to go to popular points of interest,and human movement shows a certain sequence pattern,but most existing studies on POIs recommendation do not take these factors into account.Therefore,this paper aims to improve the quality of POIs recommendation in LBSNs by using the influence of sequence fusion of geographic information,social information and popularity on check-in behavior of users.To this end,we proposed a new location recommendation method,namely GFP-LORE.First,GFP-LORE mines sequential patterns form location sequences and represents the sequential patterns as a dynamic location-location transition graph(L2TG).which contains the temporal and spatial sequence according to the check in time sequence of the same user sign in POIs.Because the new interest point depends on not only the interest point of the most recent access,but also the interest point of the earlier access in the sequence.Therefore.instead of recommending POIs to users based on first-order Markov chain,we use an efficient nth-order additive Markov chain to predict the probability that a user visit a new interest point.Finally,the sequence influence,geographical influence,friend influence and popularity of interest points were incorporated into the unified recommendation framework.In particular,the influence of friends and popularity was modeled as power rate distribution to calculate the influence,and the unified score was conducted.The score was ranked to recommend top-k new interest points.We conducted extensive experiments and a comprehensive performance evaluation of a well-known dataset(Gowalla).The experimental results indicate that GFP-LORE is superior to the accuracy of other POIs recommendation techniques,including the earliest first-order Markov chain,geo-social and their combined recommendation methods.