| As a popular research direction in the field of artificial intelligence,knowledge graph has powerful abilities in data organization and management,making it of great significance for conducting research on subject knowledge graph in education.This paper focuses on the study of an electric circuits course knowledge graph,aiming to achieve effective utilization of subject knowledge data and improve the quality of educational resources.The paper mainly works on the following three aspects:(1)Construction of the electric circuits course knowledge graph.Following the general process of constructing a knowledge graph,subject data collection is completed first.Then,the subject ontology model is constructed by analyzing the subject data.A word embedding enhanced entity recognition model is proposed,which integrates domain information and uses attention mechanism to merge static word embeddings,part-ofspeech features,and dynamic word embeddings to achieve word embedding enhancement.We construct a subject domain dataset and conduct experiments that resulted in a 3.18%improvement in the F1 score of the model,which confirms that enhanced word embedding can help improving the entity recognition task.The trained model is applied to the processing of unstructured data to achieve data expansion.Finally,structured triplets are stored and managed using the Neo4j graph database.(2)Research on question-answering algorithm based on the knowledge graph.A method for entity linking based on electric circuits course knowledge is proposed,which identifies entities in input questions,obtains entity mentions,adds them to a candidate entity set,constructs features using supervised learning,scores and sorts the elements in the set,and determines the central entity of the input question.Comparative experiments are conducted on the dataset to verify the effectiveness of the feature construction and sorting methods proposed in the paper.A method for improving the answer generation model is proposed by integrating semantic features into the input text vector representation.The input text is first enhanced with word embeddings,followed by the construction of structural features of the question,and finally the two are combined.A subject domain question-answering dataset is constructed,and experiments show that the Macro F1 score of the model after input improvement is improved by 1.96%.(3)Implementation of an electric circuits course knowledge questionanswering system.Based on the previous work of the paper,a hierarchical design for the question-answering system is proposed,with the data layer supported by the knowledge graph,the service layer containing task scheduling and algorithm model calling for the question-answering function,and the presentation layer providing an interactive interface for intelligent question-answering and knowledge point display,which has certain practical value. |
