Research On Multi-behavior Recommendation Based On Graph Neural Network

With the rapid development of Internet technology,while bringing convenience to people’s lives,it has also caused the problem of information overload.Recommendation algorithms are a key technology proposed to alleviate the problem of information overload.Currently,most recommendation algorithms use a single user behavior to predict the target behavior,while multi-behavior recommendation algorithms have received widespread attention in recent years due to their ability to fully utilize multiple user behaviors.Multi-behavior recommendation algorithms mainly solve the recommendation problem by decomposing multiple behavior matrices or introducing multi-behavior data into the learning process.However,these methods can only use low-order information for learning and cannot use high-order information.In recent years,graph neural networks have received widespread attention due to their powerful high-order information aggregation capabilities and have begun to be used in the field of recommendation algorithms,providing new ideas for solving multibehavior recommendation problems.Therefore,this paper carries out research on multi-behavior recommendation based on graph neural networks.The main research content and innovations are as follows:(1)In order to make full use of various user behavior information,we solve the problem of user and item modeling under various user behaviors through graph neural network-based embedding acquisition and parameter sharing based on multi-task learning,and model the relationships between various user behaviors based on loss optimization algorithms to solve the relationship between various user behaviors and target behaviors.We design and implement a recommendation algorithm based on graph neural networks and multi-task learning,and conduct comparative experiments on various loss optimization algorithms.The loss optimization algorithm with the best performance is used as the loss optimization algorithm of our algorithm,and it is compared with various recommendation algorithms.The experimental results show that the algorithm we proposed is significantly better than other recommendation algorithms.(2)To prevent user privacy leakage,we use federated learning to protect user privacy.We introduce a trusted third party to solve the problem of sparse client graph information in federated learning with graph neural networks and extend the graph neural network-based multi-behavior recommendation algorithm to the federated learning version to achieve the purpose of user privacy protection.Through experiments,we prove the necessity of the user-item graph construction algorithm based on a trusted third party and conduct experiments on the graph neural network multibehavior recommendation algorithm based on federated learning.The experimental results show that the algorithm can improve the recommendation effect while protecting user privacy.(3)After completing the above two research tasks,we designed and implemented a recommendation algorithm platform.Based on the analysis of the requirements of the recommendation algorithm platform,we designed the overall framework and various functional modules of the platform,implemented its system functions,and finally demonstrated the platform.