Point Of Interest Recommendation Algorithm Combined With Various Influence Factors

people can easily get many kinds of location information along with the improvement of network.The situation breeds the birth of location based social network,people can find the location he did not go but he is interested in through the LBSN,The situation also provides a good basis for the study of personalized point of interest recommendation algorithms for users.In the area of personalized point of interest recommendation,Collaborative Filtering has always been the most frequently used algorithm,This paper mainly studies in the problems of low recommendation accuracy,data sparsity and cold start in collaborative filtering algorithm and put forward two improved algorithms.First of all,this paper analyzes the current research situation of point of interest recommendation,and expounds the basic research methods of collaborative filtering.After that,we explained the advantages and disadvantages of the collaborative filtering algorithm,and then we analyzes the related issues of the point of interest recommendation research and put forward the improvement methods.Secondly,in order to solve the problem of data sparsity and recommendation accuracy of point of interest recommendation,this paper integrates geographic influence factors into point of interest recommendation.Considering the geographically cluster phenomenon of user activities,we integrate geographic influence factors into point of interest scoring prediction,and generate K current user's most interested point of interests to fill the original matrix.Then the geographic influence factors are integrated into the original matrix decomposition method in the way of the range of influence of point of interest and the active range of the user to get a more effective recommendation algorithm.Thirdly,for the problem of low degree of personalization and data sparsity in the algorithm of point of interest recommendation,we use kernel density estimation to get the distribution of user individual check-in behavior to protrude personalization.After that,we integrate social influence factors into the matrix decomposition algorithm in the way of social regular terms.Finally,we integrate similarity of point of interest into the scoring forecast and update the score to get more accurate results.Finally,we experiment the two algorithms on a truthful dataset.The experimental results show that these two improved algorithms increase the accuracy of the recommendation results and imporve the problem of data sparsity and cold start.