Research On Classroom Language Behavior Recognition Based On Text Classification

Classroom is the main place for the implementation of education in China,and the effectiveness of teaching behavior is an important factor affecting the quality of classroom.American scholar N.A.Flanders pointed out that language behavior is the main teaching behavior in the classroom,accounting for about 80%of teaching behavior.Therefore,the analysis of classroom language behavior can effectively record and study the teaching process,grasp the law and essence of the classroom,and then quantitatively or qualitatively find the problems in teaching,so that teachers can propose corresponding solutions to improve the teaching effect.Traditional analysis methods of classroom language behavior mainly include a batch of quantitative classroom teaching analysis methods represented by Flanders Interactive Analysis System,whose main research process is to manually segment and label the teaching videos.With the development of information technology in the direction of automation and intelligence,as well as the overall advancement of national education informatization,new teaching concepts and technical methods have gradually been integrated into classroom teaching,and traditional analysis methods can no longer meet the needs of teaching development.Using information technology to realize the automate classroom teaching behavior analysis and promote the improvement and development of teaching analysis methods will become an urgent need under the big wave of information.In the field of quantitative analysis of classroom language behavior,the preliminary data processing work mainly includes the segmentation,identification,classification,and coding of teacher-student behavior or information,among which classification and coding are the main links in text classification technology.At present,researches on text classification are relatively mature,and there are many applications based on content classification including news,Weibo,and web pages,while there are few related studies on classroom language behavior analysis.Based on the commonality of process between text classification and classroom language behavior analysis,this paper will use the text classification technology to classify the content of the classroom language behavior to achieve automatic recognition of classroom language behavior.The main work in the paper includes:Firstly,it sorts out the theory of classroom language behavior,summarizes the characteristics and existing problems of the existing quantitative analysis methods of classroom language behavior,and conducts a detailed review of related research on text classification,and introduce text classification technology into the analysis of classroom language behavior.Secondly,based on the study of classroom language behavior types,a set of classroom language behavior coding standards suitable for this study was developed,and the classroom teaching videos of many excellent teachers collected in the field were transcribed,organized,and classified to make a usable set of classroom discourse text,including three types of teacher discourse texts:lecture,instruction and question.Thirdly,combined with the analysis of the characteristics of classroom discourse text,this paper uses the TF-IDF algorithm to extract the statistical features of the discourse text,uses the Word2Vec model to train to obtain the word vector features,and constructs a convolutional neural network(CNN)to extract the deep learning features of the text.The three types of features are weighted and fused to explore the influencing factors of the automatic classification of classroom discourse.Finally,this paper designs 25 different discourse text representation models.By comparing and analyzing the impact of different features and different word vector weights on classification accuracy,the following conclusions are formed:(1)In the study of automatic classification of classroom discourse text,the word frequency TF is likely to increase the influence of non-characteristic words and reduce the classification accuracy,while the inverse document frequency IDF can increase the influence of feature words in class texts and improve the classification accuracy;(2)CNN can learn the in-depth features of discourse text,and has better text representation capabilities in classroom discourse classification;(3)Generally speaking,the representation capabilities of fusion features are better than a single feature,but it is not that the more fusion feature,the higher classification accuracy of discourse text;(4)Through exploration experiments,the text representation model with the best classification effect was obtained in this paper,that is,the text fusion model that fuses TF-IDF statistical features and CNN deep learning features has the highest classification accuracy,up to 88.45%.