EVSR:A Technical Framework For Extracting And Visualizing Semantic Relationships In Text-type Learning Resources

Text information extraction refers to the process of finding the target entity and the semantic relationship in a text instance with a specified entity category.Although knowledge extraction has made great progress in the field of natural language processing,there are already some knowledge extraction tools or systems that can be used for relationship extraction.However,the results extracted by most tools and systems lack usability because they do not meet the constraints of domain knowledge.Existing knowledge extraction systems cannot be directly applied to the automatic generation of knowledge maps in education.Therefore,this paper proposes a technical framework EVSR for extracting and visualizing the semantic relationship with cognitive values in text-type learning resources,which is implemented by a bidirectional long short-term memory neural network with conditional random field layer and a Piecewise Convolutional Neural Networks(PCNN).In EVSR,the BiLSTM-CRF model is used to identify semantic entities,and the PCNN model is used to conceptual entities.In EVSR,the BiLSTM-CRF model is used to extract semantic entities,and the PCNN model is used to identify semantic relationships.This paper also implemented the prototype system of EVSR,which can identify the semantic relationship entity pairs in text-type learning resources.Semantic relationship entity pairs can generate semantic link networks(SLN)through operations such as merge and inference.Finally,SLN can be visualized in the form of a concept map.This paper also compares these two models with previous researches on manual extraction methods,rule-based extraction methods,and feature-based extraction methods in terms of accuracy and adaptability.Experiments show that the accuracy rate of concept entity recognition of BiLSTM-CRF model exceeds that of our previous research,and the accuracy rate of semantic relationship recognition of PCNN model can reach a generallevel.Although rules-based and feature-based semantic relationship extraction methods can achieve higher accuracy than with deep neural network models,however,they cannot solve the problem of extracting multiple kinds of semantic relations,while the PCNN model can better handle the classification of multiple semantic relations.Compared with the visualization results of our previous research,the number and types of conceptual entities and semantic relations in the system-generated concept map are more abundant.As a knowledge visualization tool,concept maps generated from semantic relations with cognitive value can help students quickly grasp the key content and knowledge structure of the text and cultivate students' knowledge integration ability.The research also accumulated a large amount of C language corpus and C language domain small-granularity concept dictionary which are not available at present.These resources are of great significance to the construction of computer programming language knowledge base too.