Point-of-interest Recommendation Algorithm Based On Self-attention Network And Graph Neural Network

With the wide application of smart mobile devices and the rapid development of communication technology,location-based social network(LBSN)have gradually become popular.As an auxiliary tool in LBSN,POI recommendation can use the user’s historical check-in data to recommend places of interest for it,thereby improving the user’s experience.However,the current research only considers the entity connection in the user’s check-in record,ignoring the high data sparsity problem caused by limited interactive data;in addition,although some algorithms use the feature information of the check-in data,they lack the effective feature information and topology structure.Fusion results in limited recommendation performance.In view of the above limitations,this paper proposes a point of interest recommendation algorithm based on self-attention network and graph neural network.Firstly,aiming at the problem that the current point of interest recommendation algorithm is not sufficient to mine the feature information of user history check-in data,which leads to high sparsity of data,this paper proposes a point of interest recommendation algorithm based on multi-feature self-attention network.The algorithm analyzes user check-in data from different perspectives,and obtains the influence of feature information on user check-in behavior by using explicit and implicit social similarity and spatial correlation metrics.The multi-feature self-attention network model is used to capture the complex transfer relationship between different points of interest and realize the fusion of heterogeneous information,so as to recommend points of interest for users.Secondly,aiming at the problem that the current point of interest recommendation algorithm ignores the effective combination of node attribute features and interactive topological structure features in user check-in records,resulting in inaccurate recommendation,this paper proposes a point of interest recommendation algorithm based on double graph convolutional neural network.The algorithm to construct user-a user’s social graph and user interaction diagrams,interest points-traditional figure convolution neural network is utilized to extract the user’s social characteristics,based on the layered heterogeneous diagram convolution neural network learning characteristics,the interaction of the user interest points and fusion of different Angle of view features for users users are the potential vectors,and the joint point of interest in the potential vectors for predicting scores,Recommend interest points for users.Finally,the proposed algorithm was verified experimentally in two real LBSN data sets,and compared with the current mainstream recommendation algorithm,to verify that the proposed algorithm has better recommendation performance in the recommendation of interest points.