Research On Recommendation Of Point Of Interest Based On Graph Embedding

With people's pursuit of a better life,the popularity of check-in behaviors with points of interest shows people's great interest in exploring points of interest.However,the large number of points of interest and the limited time and limited energy has formed a matching problem.Therefore,it is trying to help users better find their favorite points of interest through the point of interest recommendation system,also reduce the user's decision-making cost,and improve the quality of life of the user.The interaction between users and points of interest itself can be represented by a bipartite graph structure.The graph structure has advantages in capturing the complex relationship between users and points of interest.Therefore,the research in this article focuses on general point of interest recommendation under graph embedding.This problem is divided into two parts: vector recall based on graph embedding technology and interactive function design based on search technology.The existing graph embedding technology pays more attention to node information and structural information.In the encoder design process,the input and output difference is minimized as the goal.This article follows the principle of mutual information maximization,and the encoder design purpose is to make the input and output The mutual information is maximized,so as to learn the unique representation of users and points of interest,and at the same time,the learning target is transformed into sub-layers,and the original representations of user nodes and points of interest nodes are learned through the graph neural network to obtain a user-point of interest bipartite graph The global representation and local graph representation of,and the negative samples are obtained through the destruction function,so as to use the objective function similar to the GAN network to realize the calculation of mutual information.After obtaining the vector,how to better simulate the interaction between the user and the point of interest and learn its complex relationship.Based on the design of the interaction function by the predecessors,this paper proposes a new search space to take into account the efficiency of search and recommendation The performance of the multi-layer perceptron cannot simulate the vector inner product operation well,so we first perform vector operations on the user-interest point vector,and input the result into the neural network,and finally output the prediction result,and this process is to adapt to different Types of interest points-user interaction data,we use the One-shot neural network architecture search algorithm to realize the automatic design of the neural network,avoiding the process of manually designing the interaction function.Finally,the algorithm is experimentally verified on the public real data set.The experimental results show that the graph embedding algorithm proposed in this paper on the user-point of interest is in the vector recall stage,which is lower than the recall rate and accuracy rate of the existing embedding algorithm.Certain improvement,and the interactive function searched by the One-shot algorithm can achieve improvement on the basis of graph embedding,which further expands the scope of improvement.