Research On Recommendation Algorithm Based On Graph Neural Network

With the rapid development of information technology,human society has entered the era of big data.The huge and complex data brings people more knowledge and opportunities,but also causes the problem of information overload.So how to quickly and effectively obtain useful information from the complex data becomes particularly important.The recommender system solves this problem to some extent.However,existing recommendation algorithms often assume that user interests are static when learning user preferences using user historical behavior data,so it is difficult to reflect the dynamic changes of user interests over time.Considering the time-series dependencies in user historical interaction data,this paper uses sequential recommendation modeling to capture the dynamic evolution of user interests,thereby improving the accuracy of recommendation.Further,by analyzing the current research status of sequence recommendation,this paper finds that there are still two problems in the current sequence recommendation algorithm.First of all,the current mainstream sequence recommendation algorithms still assume user behavior data to be a one-way sequence dependent sequence,ignoring the complex item conversion relationship,such as ring structure.Second,noise information is ubiquitous in the user behavior sequence,and if it is not processed,the modeled user preferences will be offset.Aiming at the above problems,this paper has carried out the following aspects of work.(1)For mainstream sequence recommendation,the user’s historical behavior data is simply assumed to be a one-way sequence-dependent sequence structure,ignoring the complex transformation relationship.This paper proposes a sequence recommendation model based on a graph recurrent attention network,which first constructs the sequence data as a directed graph,and then uses a gated graph neural network to learn the node representation based on the graph structure.Then a gated recurrent neural network is used to capture the order dependencies in the user behavior sequence,and the hidden layer vector at the last moment is used as the local interest vector.Further use the attention mechanism to assign weights to all the hidden layer vectors obtained in the previous step,let the model pay attention to the important information in the sequence,and then aggregate it into a global interest vector.Finally,the local interest vector and the global interest vector are concatenated as the user interest vector,which is further dot-producted with the candidate item vector to generate the recommendation result.(2)Aiming at the problem of noise information interference in the user behavior sequence.This paper proposes a sequential recommendation model based on improved selfattention graph pooling.The model first constructs the sequence data as graph data,and then uses the gated graph neural network to learn the node embedding vector.In order to reduce the influence of noise information,this paper further proposes an improved self-attention graph pooling method to extract the user’s core interest,and the The result of the graph readout is taken as the user’s static interest,and then,the user’s dynamic interest is modeled using a gated recurrent neural network and an attention mechanism.Finally,the recommendation is made based on the user’s static interests and dynamic interests.(3)This paper compares and evaluates the model proposed in this paper on two public datasets,and the performance has been improved to a certain extent.And further evaluate the impact of each module on the overall performance of the model through ablation experiments.