Recommendation Algorithm Based On The Spatio-Temporal Relationship In Location Social Network

Due to the popularity and development of the smart phone and GPS technology,Location Based Social Network(LBSN)has received more and more attention,such as Foursquare and Yelp.However,with the explosive growth of the number of users and items in LBSN,we are facing the problem of massive information redundancy and difficult screening,whether it is for users,merchants or researchers.Currently,the recommendation system is an effective way to solve this kind of problem.The essence of the recommendation system is an information filtering system that analyzes the user's historical behavior records to predict the user's preference for the item,and thus recommends personalized information to the user.The traditional recommendation system is mostly based on collaborative filtering technology,which utilizes users' ratings on items to generate recommendations,but due to the sparseness of the user rating data and the cold start problems,collaborative filtering technologies cannot produce accurate recommendation results.Therefore,many researchers have proposed using auxiliary information to optimize recommendations.Unlike traditional social media recommendation scenarios,in location-based social networks,there are a large number of objects with temporal and spatial attributes,such as user's text content information,user's check-in information,or the time of comment information,location's geographic information,etc.This paper takes Convolutional Matrix Factorization(ConvMF)as the basic framework and integrates temporal and space relationships to generate more accurate rating predictions.ConvMF uses the Convolutional Neural Network(CNN)technology to learn text context information from the location's review text,and then integrates CNN into the Probability Matrix Factorization(PMF).Based on the above considerations,this paper first proposes a Convolution Matrix Factorization Model with Spatial Relations(CMFSR)method.SMFSR not only considers user's rating information and location's review information,but also explores the geographical spatial relation of different locations.Tobler's First Law of Geography shows that everything is related to everything else,but near things are more related to each other.Therefore,this paper considers that the distance between locations will affect the degree of relation between them.Based on this,we constructed the location spatial relation matrix,and embed a pre-defined location spatial relation network into ConvMF by sharing item latent factor.In addition to the location spatial relationship,this paper also proposes a joint convolutional matrix factorization(JCMF)method which considers temporal relationships.The JCMF method jointly considers the item's reviews,item's relationships,user's social influence and user's reviews in a unified framework.Specifically,to explore items' relationships,if two items are rated by the same user within a short time,we assume these two items are correlated,and propose a method called Convolutional Matrix Factorization with Item Relations(CMF-I),which introduces a pre-defined item relation network into ConvMF by a shared item latent factor.To consider user's social influence,we further integrate the user's social network into CMF-I by sharing the user latent factor between user's social network and user-item rating matrix,which can be treated as a regularization term to constrain the recommendation process.Finally,to model the document contextual information of user's reviews,we exploit another CNN to learn user's content representations,and achieve our final model JCMF.Finally,we conduct extensive experiments on the real-world dataset from Yelp.The experimental results demonstrate the superiority of JCMF compared to several state-of-the-art methods in terms of Root Mean Squared Error(RMSE)and Mean Average Error(MAE).