Research On Personalized Point Of Interest Recommendation Based On LBSN

With the development of Internet technology and the popularity of intelligent mobile devices,users can easily share their favorite locations through comments,check-in,and other behaviors to record travel experiences and footprints.With the exponential growth of point-of-interest data,helping users save time,gain a good experience,and recommend suitable locations has become a research hotspot.Point of interest(POI)recommendation,as an important service in location-based social networking(LBSN),has developed rapidly.It can help users discover more interesting unknown locations and facilitate service providers to provide users with more accurate notifications or advertisements.Currently,due to the discontinuous user check-in behavior and the discrete check-in information,users face the challenges of sparse data and low recommendation accuracy.In order to provide users with more personalized recommendation and maximize the recommendation service to meet users’ needs,this paper,according to the characteristics of interest point recommendation,considers user history check-in data and POI attribute auxiliary information respectively,proposes a recommendation algorithm combining contextual information and a recommendation algorithm based on knowledge graph to alleviate the problems of data sparsity and cold start.To provide users with higher quality recommendation services.The main work of this paper includes two parts.First,in order to fully mine user access preferences,this paper proposes an interest point recommendation model(SCGM)that integrates time series,category popularity,and geographical constraints.It integrates information such as sequence,interest point category,and geographic location in the check-in history to obtain user access preferences by integrating various interest factors.The main content includes:(1)Based on the word embedding model,Obtaining a feature vector representation and a user preference vector representation of the POI from the user check-in sequence;(2)Construct a user’s virtual public access sequence based on the user’s historical check-in data,calculate the user similarity based on the virtual public access sequence similarity,and apply it to the collaborative filtering framework to obtain the user’s behavioral preference score for POI;(3)Considering the impact of geographical factors on user travel decisions,a kernel density estimation method is used to obtain user geographical preference scores for POI;(4)the preference scores of the two users are weighted and fused to obtain a POI recommendation list.The experimental results on the TKY and NYC datasets in Foursquare show that this method outperforms the other five POI recommendation methods in terms of accuracy,recall,and F1 score.Secondly,in order to alleviate data sparsity and cold start issues in the recommendation process,this paper proposes a collaborative knowledge-aware recommendation model based on knowledge maps(KGCK),which enriches the description of users and interest points by importing auxiliary information based on social networks and interest point attributes.The main content includes:(1)building a heterogeneous dissemination network,linking each POI to the corresponding entity of the knowledge map through entity links;(2)An initial seed set is obtained based on the historical check-in records of users and interest points,which is propagated along the relationship in the knowledge map to obtain an extended entity set;(3)Use attention networks to learn the feature contributions of each entity to users and interest points,and fully learn the vector representation of users and interest points;(4),the model is applied to two datasets,Foursquare and Yelp,and the experimental results show that the performance of the KGCK recommendation is significantly better than the four benchmark methods used for comparison.