Research On Graph Neural Network-based Sequence Recommendation Algorithm In Heterogeneous Information Network

Recommendation systems have been widely used in social production and our life.Recommendation technologies are also constantly pursuing extreme results and a better user experience.For the increasing scale of data,features of things are becoming more and more sparse,what's more,people are harsher about the platform's personality.The demand for better experience continues to increase,and the traditional recommendation algorithms such as machine learning bases methods and collaborative filtering methods,and sequence recommendation algorithms based on Markov processes and recurrent neural networks have not been able to suffice user's necessary.In view of the above situations,this article commences with how to effectively relieve the problem of data sparsity and how to improve the personalization ability of the recommendation algorithm.Begin with the perspective of sequential recommendation,combining the historical interaction behavior sequence of users with items to study how to improve the effectiveness and personalized requirements of the recommendation algorithm.First of all,in view of the problem of data sparsity caused by massive data,this paper proposes an end-to-end sequential recommendation algorithm based on the deep graph neural network.The method firstly models the traditional data into graph structure based on the connections between different entities.When combining different relationships between entities,it also abstracts the historical interaction sequence between users and items as graph structure data as the attribute information of entities and nodes in the graphs.The context structure information in the data is used to generate the representation vector of the node through the fusion of the graph neural network,and further predict the interaction probability between the user and the item.Secondly,in order to fully satisfy the user's demand for the personalized recommendation algorithms,our method captures the user's long-term interests and preference information.It can improve the ability to be more personalized for recommendation algorithms.Based on the graph neural network of graph structure data,this paper proposes an end-to-end deep neural network sequence recommendation algorithm,which fused with the improved Transformer encoder structure based on the recurrent neural network.The algorithm also uses the graph neural network to extract the attribute information and the context structure information of nodes.With the improved Transformer to extract the sequence information of a long sequence of users and items splice with the rich node information in the graph.The probability of interaction between users and items predicted better.It also improves the personalization ability of the recommendation algorithm,at the same time,the ability of the algorithm to face sparse data is improved furtherly.Finally,the experimental verification of the method proposed in this article.Based on multiple publicly real data sets,the method proposed in this article has been compared with the relevant baseline algorithms in a multi-angle and multi-index experimental comparison,and the experimental results have been deeply analyzed and analyzed.The experimental also analysis the reason why the method in this paper has achieved better results.