Research On Automatic Annotation Of Mathematics Test Questions Based On Multi-labe

Question automatic annotation is an important subtask of multi-label text classification in the intelligent education scenario.The research of question automatic annotation for specific subjects(e.g.,mathematics,physics)usually depends on its own subject characteristics.However,many question automatic annotation methods often can only obtain a single sentence representation of questions,which not only has the problem of semantic extraction inadequacy but also ignores the complex relationship between questions and knowledge points,making it difficult to achieve accurate question automatic annotation research.In addition,there are very few subject-specific question automatic annotation datasets.Therefore,this paper focuses on subject-specific question data to study accurate and effective question automatic annotation.The main work and innovations of this paper are as follows.(1)There are very few subject-specific test question datasets.In this paper,a dataset is constructed by collecting real mathematics questions,and 18 kinds of knowledge point labels are labeled for the test question dataset using manual annotation,with each mathematics question corresponding to multiple knowledge point labels.Therefore,the research on automatic annotation of mathematical questions based on multi-labeling aims to solve the task of multi-knowledge automatic annotation of mathematical questions.This paper is mainly based on the constructed mathematics question dataset to implement the question automatic annotation research.(2)To solve the problem of inadequate semantic extraction of mathematical questions,this paper proposes a Feature Integration Model Based on the Gating Mechanism(GFI)to achieve the task of multi-knowledge automatic annotation of mathematical questions.The model enhances the semantic representation of mathematical questions through the Gating Mechanism to integrate the important information in the local key features and contextual logical features of the question.In addition,the GFI model introduces a Mathematical Symbol Base(MSbase)to enrich the significant meaning of mathematical symbols.(3)To solve the problem of complex semantic capture of mathematical questions,based on the GFI model,this paper proposes a Multi Hop Attention Mechanism Model(MHA)to achieve the task of multi-knowledge automatic annotation of mathematical questions.The model focuses on different important information in mathematical questions through the Multi Hop Attention Mechanism.In addition,due to the advantages and efficiency of the A Lite BERT(ALBERT)pre-trained model,ALBERT is still used in the MHA model to accomplish the word embedding task in the research.(4)On the constructed mathematical question dataset,this paper conducts experimental validation of the proposed GFI model and MHA model.Experimental results finally show that both the GFI model and the MHA model can achieve better annotation accuracy,and the MHA model obtains better results than the GFI model.