Research On Sequential Recommendation Algorithm Based On Attention Mechanism

With the deepening of information technology,the amount of data on the Internet has increased dramatically.The problem of information overload in the field of the Internet is becoming increasingly serious.In order to alleviate this problem,researchers have proposed recommendation algorithms to predict user preferences.This paper studies sequential recommendation algorithms,which models historical behavior to predict future behavior.The current research on sequential recommendation algorithms uses many new technologies,such as neural networks,attention mechanisms,heterogeneous graphs.These algorithms have made considerable progress.However,the following problems still exist in the research of sequential recommendation algorithms:(1)The algorithm using heterogeneous graphs is not clear about the logical division of user behavior;(2)Existing sequential algorithms do not attach importance to the association of auxiliary information;(3)The basic operation unit of the existing sequential recommendation calculation is single.To solve these problems,the research content of this paper is summarized as follows:(1)To solve the problem that sequential algorithms using heterogeneous graph modeling do not effectively utilize differences in user behavior patterns,this paper proposes a Session Recommendation Algorithm Based on Heterogeneous Hypergraph and Composite Self Attention Mechanism(HHCSA).The algorithm utilizes the differences in user behavior patterns to partition behavior.The algorithm uses corresponding learning blocks to obtain behavioral features.Session information and auxiliary information are modeled as heterogeneous hypergraph structures.Hypergraph information aggregation algorithms aggregate items and auxiliary information features in heterogeneous hypergraph.The aggregated features are input into impulsive behavior and rational behavior learning algorithms to obtain behavioral representations.The gating mechanism combines behavioral characterization to predict results.Experimental results show that HHCSA algorithm can effectively model sequential information.And user intent partitioning can effectively improve performance.(2)To solve the problem that auxiliary information association is ignored by existing algorithms and the granularity disappears after information fusion,this paper proposes an Auxiliary Information Feature Fusion for Sequential Recommendation(AIFF).The algorithm uses attention mechanisms to extract dependencies between auxiliary information.The auxiliary information representation serves as the query value and key value for attention mechanisms.Sequential information as the real value of attention.Attention mechanism and neural network learn the correlation between sequential information and auxiliary information to complete prediction tasks.Sequential information and auxiliary information are input into attention mechanisms and neural networks to generate predictive information.Comparative experiments show that feature fusion components can effectively learn auxiliary information.Extensive experiments show that the auxiliary information fusion algorithm of the AIFF algorithm can be extended to algorithms using attention mechanisms.Moreover,the algorithm prediction performance of the extended algorithm can be improved.(3)In order to solve the problem of limited information granularity caused by using too simple modeling objects,this paper proposes a Sequential Recommendation Algorithm based on the Item Cluster Learning Unit(ICLU).This paper models the sequence as a cluster pattern transition diagram.Clusters are divide into jump clusters and continuous clusters.The algorithm introduces a cluster learning unit to extract cluster features.Gating mechanism combines cluster characteristics to obtain prediction results.The experiment shows that the strategy of item cluster modeling has significant advantages compared to the original scheme.