In recent years,the research on recommendation algorithm has emerged in an endless stream,and the application of intelligent recommendation algorithm has been extended to many fields.In the field of education,teaching resource recommendation algorithm has become an important research topic.Although many recommendation models with excellent recommendation effect have been proposed in the existing research,there are still some limitations.First of all,most recommendation models based on knowledge graphs only process the local context information of nodes,and do not make full use of the non-local context information of nodes.Secondly,most session-based recommendation models only conduct item or behavior sequence representation learning for one type of behavior in historical session records,without considering the influence of other types of learners’ behaviors on recommendation results.Finally,most of the existing recommendation models only extract features from resource structure information or learners’ historical session information to aggregate and update project information,and few of them are used together in recommendation tasks.Therefore,this paper explores the resource graph and behavior graph more fully,takes the context information in the resource graph and the historical behavior information of learners into consideration,proposes a new personalized recommendation model of teaching resources,the DB-CGAT model,learns the item representation from a global perspective,and finally realizes the multidimensional preference personalized recommendation.The specific research content of this paper is as follows:(1)In terms of the structure of teaching resources themselves,this paper proposes a method of contextual processing of knowledge graph of teaching resources to fully extract contextual information in knowledge graph of teaching resources themselves.In this method,the local and non-local neighborhood nodes of each node on the knowledge graph are obtained by using different aggregation strategies to obtain the local and non-local context embedding,and the two are aggregated again to obtain the global context embedding of each teaching resource node.(2)In the aspect of learners’ historical session,this paper proposes the dual behavior aggregation method to fully explore and utilize the interactive behavior information in learners’ historical session.This method aims at two kinds of behavior sequences in learners’ historical session,namely target behavior and auxiliary behavior.Through a series of operations of graph attention network and graph neural network,the project representation and behavior sequence representation in historical session are updated,and learners’ preference representation in session is further obtained.(3)This paper proposes a personalized recommendation model for teaching resources(DB-CGAT),which combines the knowledge graph context processing method with the dual behavior aggregation method for the recommendation task.In the model,all learners’ historical interaction items are corresponding to the knowledge graph of teaching resources,the context processing method of knowledge graph is used as the preprocessing of the dual behavior aggregation method,and the global context embedding of teaching resource items on the knowledge graph of teaching resources is used as the initial embedding of the item in the dual behavior aggregation method.Participate in the information aggregation among nodes in the subsequent behavior graph and multidimensional preference personalized recommendation,so as to further enhance the reliability of teaching resource recommendation and realize the personalized recommendation.(4)The effectiveness of DB-CGAT model is verified by experiments.In this study,experiments were conducted on two real data sets,and the experimental results show that the DB-CGAT model proposed in this paper is better than the traditional methods in terms of teaching resource recommendation,and has good performance. |