Point-of-Interest Recommendation Integrating Social Networks And Visual Contents

Point-of-Interest(POI)recommendations are part of the recommendation system and are deeply rooted in people's daily lives.With the continuous development of LBSN technology,the recommended application of interest points has accumulated a large number of users,resulting in a large number of points of interest data.This paper mainly focuses on the problem of inaccurate recommendation and low recommendation accuracy caused by sparseness of interest point data in the recommendation field of interest points,and proposes different methods of model fusion.Firstly,this paper proposes a method DTPOI for the conversion of the commodity recommendation domain algorithm to the POI recommendation domain.It is different from the traditional recommendation model directly using the score of the rate,but the user rate information is redefined;this article is based on the geographical location information and the tag information constructs the distance factor and the tag factor respectively,and combines the user scores to calculate the interest review score information matrix to participate in the matrix decomposition.Through the experiments on the Yelp real dataset,the DTPOI algorithm has an average improvement of 47.765%and 22.77% compared with the PMF,SoRec,TrustMF,and TrustSVD algorithms in the MAE and RMSE evaluations.The experimental results show that the DTPOI algorithm improves the accuracy of the POI recommendation.Secondly,this paper proposes a POI recommendation algorithm VPOI that fuses image content.In this paper,the VGG16 model of deep convolutional network model is modified to make feature extraction of the image of POIs.At the same time,the feature vector of POI image which can be merged by probability matrix is constructed to form the feature image of the POI image.Through the experiments on the Yelp real dataset,the VPOI algorithm has an average increase of 19.535% and 8.705% compared with the PMF,SoRec,TrustMF,and TrustSVD algorithms in the MAE and RMSE evaluations.The experimental results show that the VPOI algorithm improves the accuracy of the POI recommendation.Finally,this paper explores the social relationship in the field of interestrecommendation,and based on the social matrix decomposition model,the previous exploration work is modeled,and POI recommendation algorithm SVPOI is proposed to integrate social network and image content.The SVPOI algorithm utilizes user rating information,geographic location information,tag classification information,user social relationship information,and POI image information for POI recommendation.Experiments were carried out under the Yelp real data set.The experimental results show that the MAE and RMSE values of the SVPOI algorithm are increased by 5.5% and7.82%,respectively,compared with the optimal algorithm TrustSVD under the DTPOI method.The experimental results show that SVPOI has better recommendations.Accuracy.