Research On Recommendation Algorithm Based On Significance And Similarity Score Of Interest Points

Location-based Social Networks(LBSN)pro vides users with many important location-aware services.Points of Interest(POI)recommendation is one of the most important services.The point of interest recommendation recommends a new and attractive place to the user by mining the user's check-in behavior.Because of its huge commercial value and application prospects,it has attracted wide attention from academia and industry.Most of the existing recommendation algorithms combine one or more of the four main factors of geography,time,social interaction and content to improve the recommendation effect,and the degree of consideration for the interest point itself is not high.From the point of view of interest points,this paper makes full use of the sign-in behavior in LBSN,and explores the importance of interest points in the whole social network and the similar characteristics between various points of interest to improve the quality of recommendation.Firstly,according to the phenomenon of geographic aggregation,we analyze the activity characteristics of users in the real checked-in data set,Foursquare,and filter dirty data.According to the fact that the user's activity range and interest point affect the limited spread range,we construct the user's activity range vector and interest point influence range vector.Aiming at the problem that the influence of all interest points is treated equally when the traditional recommendation algorithm models the influence of interest points,a recommendation algorithm based on the importance of interest points is proposed.The algorithm defines the calculation method of the importance degree,and considers the importance of candidate interest points in the prediction scoring stage,so as to improve the recommendation quality.Secondly,using the user's check-in behavior to mine the similarity between interest points,a recommendation algorithm based on the similarity score of interest points is proposed.The feature vector of each interest point is trained by Word2 vec principle,and the similarity between interest points is calculated.The calculation method of the similarity score is considered in the prediction and scoring stage,and the similarity scores of the candidate interest points and the user's visited interest points are considered to improve the recommendation effect.Finally,the experiment is carried out on the real check-in data set-Foursquare,and compared with the current advanced algorithms,the effectiveness of the proposed algorithm is verified.