Research On Entity Relation Extraction Model Of Middle School Mathematics Subject Knowledge Graph Based On Graph Representation Learning

Intelligent education integrates a large amount of data,content,tools and other resources by using information technology,artificial intelligence and other advanced technologies to achieve personalization,precision and efficiency in education.As an effective semantic representation tool,knowledge graphs can transform massive data into structured knowledge,thus reorganizing and integrating educational resources.Relation extraction can extract the knowledge and information used to build the knowledge graph from the text,and is a crucial process of the knowledge graph construction.Entities and relations in educational texts are usually distributed across sentences,and relations between entities need to be inferred from the global information of the document.In addition,current document-level relation extraction datasets are limited in domain coverage in terms of size and quality,and there is a lack of data in the education domain.As a key link in the contruction of intelligent education,knowledge graphs cannot combine data and knowledge without the data support of educational relationship extraction,thereby slowing down the progress of intelligent education.To deal with these problems,this thesis conducts research on document-level relation extraction model for the education domain.The main research work can be summarized into the following two aspects.(1)In this thesis,we propose a document-level relation extraction model fused with educational text features.The model models the entities and their relations in a document by constructing a document graph fused with educational text features,and employs graph neural networks to perform update and aggregation operations on the nodes in the document graph.The model uses graph path inference to capture indirect relations between entities across sentences in a document,and finally uses a classifier to perform prediction of relations between pairs of entities.(2)In order to validate the effectiveness of the model,based on the structural and modular features of the educational field texts,we construct a document-level relation extraction dataset for middle school mathematics subjects in this thesis,referred to as the Math dataset.With the practical application scenarios,this thesis selects educational texts such as middle school mathematics textbooks and teaching designs as data sources,and invites experts in the field to participate in the data annotation work,and finally obtains a document-level relation extraction dataset that can cover all relevant knowledge points of middle school mathematics chapters.In this thesis,the proposed model is experimentally validated on the Math dataset and compared with other document-level relation extraction models.The experiments show that the model in this thesis performs well in terms of precision rate,recall rate and F1 value,which proves the effectiveness of the model.The research in this thesis provides practical references for the development of document-level relation extraction in education,and supports the construction of subject knowledge graphs and the digital and intelligent management of educational resources.