Research On The Automatic Evaluation Of Composition Of Primary School Based On Vocabulary And Sentence Level

In the course of Chinese language learning,how to cultivate students' interest in writing and improve their writing ability has always been a hot issue of concern to educators.Traditional composition scoring generally requires manual scoring by the teacher,which not only consumes a lot of manpower,but also is subject to the personal subjective influence of the evaluator.With the rapid development of artificial intelligence-related technologies such as natural language processing,the use of computers to automatically evaluate composition and assist teaching has become a research hotspot.This thesis takes primary school Chinese composition as the research object,and carries out a series of researches on automatic evaluation methods for primary school composition.By combining the traditional composition evaluation standards and the characteristics of the development level of primary school students' writing ability,this thesis proposes a method of primary school composition through vocabulary and sentence level Method of evaluation elementary school.The main work of this research is as follows:(1)Through related research and theoretical analysis,this thesis describes the principles and characteristics of key technologies that may be involved in the automatic evaluation process of the composition.Finally,based on its performance and feasibility analysis,the research route of this thesis is clarified.(2)Based on vocabulary use as the research point,this thesis establishes a lexical-based multiple linear regression model and equation.Firstly,the correlation between vocabulary richness,lexical complexity,and vocabulary frequency profile and composition scores was studied.The vocabulary scores used in the thesis were weighted to obtain an equivalent measure of composition score.Multivariate linear regression model,and finally optimized the vocabulary of the proportion of complex words,the number of most commonly used words,the number of words and the control of text length TTR(random extraction)as the four factors to predict composition performance.(3)This thesis studies the method of automatic extraction of sentence levels in primary school composition.First,by analyzing the characteristics of sentence hierarchy,the criteria for sentence hierarchy classification were determined.Then the sentences are classified according to the background of related disciplines,and this is used as a training set to construct a sentence level classification model by using bidirectional LSTM.The experimental results verify the effectiveness of the model.The classification accuracy of A-level sentences is 88.32%,the classification accuracy of B-level sentences is about 84.27%,and the accuracy of C-level sentences is 71.73%.(4)Based on the above techniques,this thesis constructs an automatic evaluator of primary school composition based on vocabulary and sentence levels,and designs related comments based on various scoring features.Finally,the data set collected in the previous period is tested and compared with the experimental results of related research to verify the correctness of the research idea.