Information Fused POI Recommendation Algorithm Based On Matrix Factorization And Tensor Decomposition

The popularity of the network and mobile devices has promoted the widespread use of Location Based Social Networks(LBSN),which has brought great development prospects for personalized Point-of-Interest recommendation systems.At the same time,the POI recommendation system provides users with more plentiful travel plans,and it also brings more business opportunities for companies and merchants.Collaborative filtering algorithm has been the most widely used algorithm in the recommendation system,but it has the problems of insufficient accuracy,sparse data,poor scalability,and cold start.In addition,considering that user travel is affected by many factors,it is necessary to establish and explore more relationships within factors.Therefore,this paper proposes an information fused POI recommendation algorithm based on matrix factorization and tensor decomposition.In this paper,matrix factorization and tensor factorization models are used to fully combine each impact factor to provide users with POI recommendations that match their personalities.Firstly,the research status of POI recommendation system at home and abroad is analyzed,and its development history from collaborative filtering model to matrix factorization model to tensor decomposition model is introduced.The classification of collaborative filtering algorithms and the advantages and disadvantages of the algorithms in each classification are described in detail.The concepts and classifications of currently popular matrix factorization algorithms,and the definition and decomposition methods of tensor factorization models are also described in detail.Then,in order to alleviate data sparseness and improve user implicit feedback,this paper first proposes a POI recommendation algorithm based on social relationship exploration.We use the rich contextual information(such as time,location,friends,etc.)in the check-in records to obtain the regional influence,distance influence,and friend influence.Then potential points of interest into the matrix factorization model is obtained and the access probability is predicted.The top-N points-of-interest are recommended to users.Among them,due to the consideration of the influence of friends,we deeply explored the user's social circles and proposed three different definitions of friend relationships: online social friends,regional friends,and rest friends.The geographical location of POI and the social relationship of users are used for most of the current POI recommendation.Usually,two or three influencing factors are considered at the same time.The incomplete utilization of influencing factors makes it difficult to improve accuracy.In order to strengthen the relevance and comprehensiveness of influencing factors,a general model of POI recommendation based on tensor decomposition for multi-dimensional information fusion is proposed.The model is divided into two parts: user portrait and location label.We use social relationships,POI categories and time factors to construct a third-order tensor and define user resident locations for user portraits.Geographic factors and popularity are used to influence the construction of location labels.Combining the two to assign influence factor weights,finally TOP-N POIs for each user are recommended.Finally,experimental results on Foursquare data set show that the proposed algorithm not only has flexible scalability,low complexity,ability of solving the cold start problem and realization of local and remote recommendation,but also has a significant improvement in accuracy compared with the current popular point of interest recommendation algorithms.