Research On Academic Resources Recommendation Algorithms Based On Knowledge Graphs

In the era of booming Internet technology,the number of academic resources continues to grow.In the face of massive academic information,the recommendation system is one of the most efficient information filtering methods,which effectively alleviates information overload and improves resource utilization.Academic resources are multi-type and heterogeneous,and users have diverse and potential interests and tendencies.Traditional recommendation methods are susceptible to data sparsity and less diversity in recommendation results.The current effective mitigation method is to fuse side information to recommendation systems.As a research hotspot in recent years,the recommendation methods with knowledge graph can effectively organize the rich and relevant knowledge in academic resources,which could be naturally integrated into a recommendation system,and improve the accuracy and diversity of results.In response to the above problems,with the advantages of knowledge graphs,this paper takes paper recommendation as the entry point.This paper focuses on modeling paper with knowledge graphs,user reading preference and personalized papers recommendation,which makes personalized academic papers more accurately.The research contents and main contributions in this article are as follows:Firstly,this paper proposes a paper recommendation framework KIRec to study how knowledge graphs are incorporated into paper recommendation,attempting to improve the recommendation performance.Regarding user-item interaction and metadata into entities and relationships to build a collaborative knowledge graph based on user-item bipartite graphs.Then,an attention network could distinguish the contribution of user interaction records to his preferences,and ultimately guide to fuse the user's node neighborhoods.To explore the user's potential and diverse interests further in knowledge graphs on the KIRec framework,this paper proposes a recommendation algorithm,GNPR,based on the users' explicit and implicit reading preferences.This time,we construct a concept-level knowledge graph and explore the implicit interest of users based on the improved GCN model.In addition,to integrate the explicit interest embodied in the text features,and propose a two-layer self-attention mechanism to obtain the internal global characteristics of the text to capture more explicit user interest.Secondly,based on the open paper recommendation datasets and real academic recommendation application log data,this paper designs and implements a series of experiments to verify the feasibility of KIPRec and GNPR,comparring the advantages and disadvantages with baseline methods.The experiments show that the method modeling user interest with knowledge graph and text features can improve the effect of paper recommendation.At the same time,the ablation experiment verifies that every part of the models is necessary,such as the attention mechanism module,knowledge graph representation learning module,and user-item interaction modeling.Finally,this paper designs and implements a multi-type academic resource recommendation system,"academic headlines",to explore the scalability of KIRec and GNPR in an engineering application.The recommendation system applies the above methods to various resource recommendation scenarios,such as patents,books and news,which can meet the needs of users more than the content-based and graph-based methods and illustrate the advantages of the recommendation system with knowledge graphs.