Study On Automatic Evaluation Of Pupils' Composition Fluency

The fluency of composition is an important indicator in the evaluation of composition,which reflects the normative nature of the author’s language and the degree of smooth of the article.The evaluation of essay fluency is of great significance to researchers and teachers who are committed to improving students’writing skills and the effectiveness of essay evaluation.Since the group of primary school students are mostly in the language learning stage,the quality of their composition depends to a large extent on the fluency of the language.Therefore,the effective evaluation of fluency is more important in the composition of primary school students.However,the traditional artificial form of composition evaluation has the disadvantages of large workload,strong subjectivity and long feedback period.Therefore,more and more scholars have devoted themselves to the related research of automated essay evaluation,hoping to use automated means to complete the work of essay evaluation.The automated evaluation of fluency is an indispensable part of automated essay evaluation system,and it is also one of the core contents of automated essay evaluation research,which has very important research significance.The paper selects the Chinese composition of primary school students as the research object.Based on the idea of automated essay evaluation,the machine learning method is used to construct the classification model,so as to realize the classification of composition according to the fluency level,that is,to realize the automated evaluation of the fluency of primary school students’ composition.Based on the clarification of the essay fluency and the concept of automatic essay evaluation and related research,the paper analyzes and selects a series of quantitative features that reflect the fluency of the composition from the total essay,paragraph,sentence,phrase,vocabulary and grammatical errors.Thus,the overall composition fluency feature set is established,and the extraction methods of each feature are explained.Then the paper follows the standard machine learning paradigm to build and experiment with the model.After the preliminary work of data preprocessing and feature extraction with LTP and CRIE tools,the study uses the Information Gain and Gain Ratio to select the features,and compared the performance differences of the three single models of logistic regression,decision tree and support vector machine before and after feature selection.Subsequently,the paper constructs three integrated classifiers,LMT,SimpleLogistic and RSM.Finally,the paper compares the overall classification accuracy of the composition fluency evaluation model based on the above six algorithms,and finds the optimal prediction model is SimpleLogistic which combines logistic regression and decision tree,and its classification accuracy rate is over 85%.The experimental results demonstrate that the model constructed in the paper has a good effect on distinguishing the fluency of primary school students’ composition.However,due to the lack of depth and breadth of the feature indicators extracted in the study,and the scope of the training set is relatively small,so further improvement and perfection are needed.