Research On Scholar Recommendation Algorithm Based On Knowledge Graph

In the big data environment,traditional recommendation methods generally suffer from data sparsity and cold start problems.As a large-scale heterogeneous semantic network,knowledge graph has the characteristics of rich semantics,good quality and friendly structure.Many studies have introduced knowledge graphs into the research of recommender systems,and studied interpretable recommendations based on knowledge graphs.However,existing studies often perform poor when faced with sparsely related graphs,and lack a certain degree of interpretability.At the same time,as a link between knowledge utilization and creation,the academic field is flooded with a large amount of knowledge,and a recommender system needs to be introduced to solve the problem of information overload.Therefore,taking scholars as the main subject,analyzing the scholar’s field of study,work units,and relationships between scholars,using scholars’ knowledge to build a knowledge graph,and recommending relevant scholars for users based on the knowledge graph.It has certain guiding significance and practical application value for grasping the academic trends,the frontier of science and technology,the development of research work and the introduction of talents.The main work as follows:(1)Scholar knowledge collection and structured representation.Firstly,for the semi-structured data of the SCHOLAT,the required entities,relationships and attributes are extracted from it.For unstructured data,the entity extraction model BERT-Bi LSTM-CRF is used to extract the entities in the text.Secondly,according to the obtained two sets of data,data preprocessing is performed and triples are constructed,and then the entities are aligned using the IPTrans E.Finally,a scholar knowledge graph is constructed based on triplet data,and Neo4 j is used to store knowledge.(2)Aiming at the problem of poor performance of knowledge graph-based recommendation methods in the face of relational sparse graphs and lack of interpretability,a recommendation method for scholars based on high-order propagation of knowledge graphs(Ho PKG)is proposed.First of all,we use the Tran Sparse to embed the scholar’s knowledge into the vector space to obtain the corresponding vector representation.Then,according to the characteristics of the scholar’s knowledge graph,the attention mechanism is used to calculate the attention scores between different nodes,thereby distinguishing the importance of different entities to the target entity.At the same time,based on the idea of ? ? inductive learning,the high-order propagation mechanism is used to obtain high-order semantic information in the knowledge graph,and then generate richer entity representations.On this basis,a new entity aggregator is proposed to further refine the entity representation to capture the potential interests of users.After experimental analysis,the method has certain advantages in recommendation effect and interpretability.(3)Based on the proposed recommendation algorithm,a visual verification system for the academic field is designed and implemented.According to the user’s interaction history and retrieval content,it can accurately recommend scholars in related fields for users.In addition,the system can offer a visualized scholar relationship discovery,which is designed to provide interpretable basis for recommendation.