Research On Point-of-Interterst Recommendation Algorithm Based On Hetergenous Network Embedding

Point-of-interest recommendation as an effective method to solve the information overload in Location-Based Social Network(LBSN),The accuracy of its recommendation and the robustness of the model directly affect the user's trust in the recommendation system.Therefore,point-of-interest recommendation has important research significance.In order to effectively alleviate data sparseness,many researchers have integrated the check-in behavior information,such as time factors,geographic distance,popularity of point-of-interest.However,these studies rarely involve seasonal information.Not only the user's interest is affected by their own behavior,but also to a certain extent by the surrounding friends.However,the existing research seldom pays attention to the joint influence of user social relationship and check-in behavior information on point-of-interest recommendation.In response to the above problems,this article incorporates seasonal information into user check-in behavior information,and combines user check-in behavior information with user social relationships,and conducts in-depth research on point of interest recommendation.Firstly,in view of the influence of check-in behavior information on the recommendation of points of interest,as well as the problem of ignoring seasonal factors that affect personalized recommendation,this paper proposes a POI recommendation algorithm based on multiple heterogeneous networks.Our algorithm divides the level of network according to the check-in season,and at the same time combines spatial information,the popularity of POI,and the user's seasonal preference for POI and other factors to customize the weighting strategy,thereby constructing a multi-information fusion multi-weighted heterogeneous network.Subsequently,the node sequence is generated according to the custom meta-path random walk.The self-attention mechanism is used to capture the influence of the seasons on the nodes in different levels of network,and the feature vector representation is obtained through the heterogeneous Skip-Gram model.Finally,POIs are recommended based on the user's preference score ranking.Secondly,in order to combine the common influence of user social relationships and check-in behavior information on point-of-interest recommendation,improve the stability of the recommendation algorithm,this paper proposes a weighted heterogeneous network embedding POI recommendation algorithm.The algorithm uses user social information and user check-in behavior information to establish the association between users,locations,and time,thereby constructing a weighted heterogeneous network of multi-information fusion.According to randomly generated meta-paths,the association relationship between different types of nodes is enriched and retained.Generate node sequence with partial wandering on the meta-path.The feature vector representation is obtained through the heterogeneous Skip-Gram model,and finally the obtained feature vector is used to recommend POIs for users.Finally,the algorithm proposed in this paper has been experimentally verified on the Foursquare and Gowalla datasets,and compared with existing POI recommendation algorithms.The experimental results verify the effectiveness of our algorithm.