Research On Graph Neural Network Recommendation Algorithm With Attention Mechanism

With the rapid increase in the amount of information on the Internet,the problem of information overload has become an important factor restricting the development of the Internet.As an effective method to solve this problem,recommender systems have received more and more attention and research in industry and academia.As an important data structure,graphs can represent complex relationships between a set of objects.Most of the data in the recommender system has a graph structure in nature.Applying the graph neural network to the recommender system can more effectively understand the user’s preferences and needs from various data.Therefore,the recommendation system based on graph neural network has become an important research direction in the field of recommendation system.This thesis proposes two recommendation algorithms based on graph neural networks: graph neural network algorithm MGRU fused with multi-head attention mechanism and graph neural network algorithm GNNLSR with long-term and short-term preference fusion.In the graph neural network recommendation system,the user’s interests and hobbies are affected by their own historical behavior,social network and other aspects,showing a dynamic trend.However,how to combine the user’s social network information and time series interests to extract effective information in the recommender system is a difficult problem.Aiming at the above problems,the MGRU algorithm utilizes the timing information of the selected memory and forgetting nodes of the gated recurrent unit to enhance the abstraction ability of the graph neural network during node iteration.Then,the attention memory network is used to obtain the influence of friends on users in different aspects,and the multi-head attention mechanism is used to adjust the influence of friends.Experiments were carried out on the Ciao and Epionions datasets using root mean square error and mean absolute error as evaluation indicators,and the results showed that the algorithm improved the accuracy of the recommendation system.In the current recommendation methods based on graph neural network,many scholars construct the interaction information of users and items into a graph,and then obtain the embedded representation of users and items by aggregating and updating nodes on the graph.However,most current models based on graph neural networks usually only consider a certain aspect of usergenerated short-term preferences and long-term preferences.The essence of user preferences is the combination of short-term preferences and long-term preferences.Long-term preferences record a user’s long-term habits,while short-term preferences are new interests generated over time.Therefore,this thesis proposes a graph neural network algorithm GNNLSR that integrates long-term and shortterm preferences.In addition,the feature extraction of items is also integrated in this algorithm,so that the model can generate better recommendation results based on capturing user’s long-term and short-term preferences and combining item features.