An Automatic Feedback Model With Multidimensional Linguistic Features Of Chinese Compositions For Primary Schools

Chinese writing has always occupied a core position in the whole process of Chinese teaching,and it is one of important teaching contents for cultivating students'comprehensive literacy.Because assessing compositions is time-consuming and laborious for teachers in traditional Chinese writing instruction,students cannot get the feedback results in time;in addition,due to fatigue or personal emotions,teache's grading is relatively subjective.In this vein,it is quite important to investigate how to use computers to solve the problems of manually evaluating compositions in a traditional way.With the rapid development of technologies,such as natural language processing and machine learning,it becomes more feasible to use computers to automatically evaluate compositions.Therefore,the thesis aims to develop an automatic feedback model with multidimensional linguistic features of Chinese compositions for primary schools.It is hoped that students can be assisted in evaluating and revising their compositions by means of automatic feedback.Besides,this study also conducts an empirical study,which involves six graders in an elementary school in Wuhan city as an example.The thesis includes three main research contributions,described as follows.First,this study proposes 120 linguistic features corresponding to the evaluation of Chinese essays for primary schools from the four dimensions of word foundation,text structure,language expression and emotional theme.This study also adopts natural language processing techniques and extracts the features from a corpus with the 4,193 collected compositions according to shallow and deep language features.The feature vector space model(F-VSM)and support vector machine(SVM)were used to verify the reliability of features.The result shows that the average classification accuracy rate based on the shallow feature F-VSM algorithm is 63.67%,the average classification accuracy rate of the SVM algorithm is 72.03%;the average classification accuracy rate based on the deep feature F-VSM algorithm is 81.33%,the average of the SVM algorithm The classification accuracy rate is 84.89%;the average classification accuracy rate based on the hybrid feature F-VSM algorithm is 76.41%,and the average classification accuracy rate of the SVM algorithm is 90.64%.The results suggest that both shallow and deep linguistic features should be considered for the selection of composition features.Second,an automatic feedback mechanism is proposed based on the linguistic features in the four dimensions.This study also constructs an automatic feedback model on the basis of an automatic essay evaluation system.The model includes single indicator visualization and overall composition index analysis.The function modules of the system are designed and implemented accordingly.Thirdly,the study conducts an experiment,in which six graders from a primary school in Wuhan participate to test the effectiveness of the automatic feedback model in the automatic essay evaluation system.The analysis on the overall modification of the compositions show that the composition quality of the experimental group improves significantly better than that the control group in general.The analysis on the linguistic features further indicates that the composition quality of the experimental group is significantly better than that of control group on the dimension of chapter structure.The results suggest that the proposed automatic feedback method may help students improve their writings,especially the structure of their compositions.