Research On Continuous POI Recommendation Algorithm Based On Space-time, Social And Image

In recent years,as location-based social network services(LBSNs)have become more and more popular in people's lives,POI(Point-of-Interest)recommendation has become a valuable research topic.The goal of the POI recommendation is to recommend sites that have not previously been visited but may be of interest to users,which is of high research value for both users and merchants.Successive POI(point-of-interest)recommendation in location-based social networks(LBSNs)is an important research topic in the field of data mining.It can recommend interest to users based on the user's current check-in information and the use of links between sign-in points of interest.The point can be timely feedback to the user in the next period of time the user may like.In current research work,the integration of time factor,social relationship,and image information is usually not considered to improve the system's recommendation performance.To solve this problem,this paper proposes a variety of continuous POI recommendation models that incorporate social relationships,spatio-temporal information,and images.These models are based on an ordered learning framework and jointly model users,time,points of interest,social relationships,and image information.The experimental results on multiple data sets show that the integration of time factors,social relationship factors and image information can effectively improve the performance of continuous POI recommendations.This paper mainly studies the POI recommendation based on,social and image information.The main contents include:First of all,in some current research work,the use of a combination of time and geographic factors is not usually considered to improve the recommended performance of continuous POI.Therefore,this paper combines time information and geographic information into continuous POI recommendation from two perspectives.The algorithm proposes continuous POI recommendation models PMRE-GT1 and PMRE-GT2 based on metric embedded framework fusion time and geographic factors.Previous time-aware POI recommendation models did not consider how to reasonably divide time periods to improve recommendation performance.To solve this problem,this paper For the first time,a reasonable segmentation time period using clustering method is proposed,and the user's time parameters are learned by adding time eigenvectors.In order to prevent the phenomenon of over-fitting,this article adds a time regularization term to the objective function.Secondly,in this paper we study successive POI recommendation algorithms based on metrics embedding frames into social relationships,and proposes PMRE-GS1 and PMRE-GS2 models.Calculate the similarity between the signing behaviors of users and friends,and add them to the social relations regular items,and use the preferences of the users' social friends to constrain the user's preference vectors.Moreover,this paper integrates the spatio-temporal information and social relationship factors into the metric embedded framework at the same time.Thirdly,this paper studies the successive POI recommendation algorithm based on image information and geographic information,using the CNN deep learning framework to extract the user's image feature vector,and integrates it into the Bayesian personalization sorting framework and propose a VSPOI model.Finally,we experimented with the multiple models presented in this article on multiple real data sets and compared the performance of successive POI recommendations.The experimental results show that the recommended performances of the time-awareness model,social-relationship model and VSPPI model presented in this paper are better than previous work.It shows that integrating time,social information and geographic factors into successive POI model can effectively improve continuous POI recommended performance.