基于语言学特征的小学生作文流畅性自动评价  

Research on Automatic Evaluation of the Fluency of Elementary School Students’ Composition Based on Linguistic Features

作　　者：吴恩慈 田俊华[1] Wu Enci;Tian Junhua

机构地区：[1]南京师范大学教育科学学院,南京210097

出　　处：《教育测量与评价》2020年第3期41-50,64,共11页Educational Measurement and Evaluation

基　　金：江苏省高校优势学科建设工程(PAPD)资助项目“南京师范大学教育学优势学科”研究成果

摘　　要：基于作文自动评价的研究思路,采用机器学习的方法对小学生作文流畅性进行自动评价。首先,从作文的总篇、段落、句子、短语、词汇和语法错误层面分析并抽取了一系列能够反映作文流畅性的语言学计量特征;其次,基于逻辑回归、决策树和支持向量机三种经典模型以及逻辑模型树、SimpleLogistic和随机子空间三种集成模型,训练了六个流畅性分类器进行实验。实验结果表明,文中选取的17个特征项对于作文流畅性具有较好的区分度,训练的模型能够根据流畅程度对作文进行较为准确的分类,集成了逻辑回归和决策树模型的SimpleLogistic分类器表现最佳,其分类准确率达到了85.2%。Based on the research ideas of automatic evaluation of composition,machine learning method is used to evaluate the fluency of elementary students’composition automatically.Firstly,a series of linguistic econometric features that can reflect the fluency of composition are analyzed and extracted from the total essay,paragraph,sentence,phrase,vocabulary and grammatical errors.Secondly,based on three classical models of logistic regression,decision tree and SMO and three integrated models of LMT,SimpleLogistic and RSM,six fluency classifiers were trained.The experimental results show that the 17 feature items selected have a good distinction for composition fluency,and the trained model can classify the composition accurately according to the fluency.The SimpleLogistic classifier with integrated logistic regression and decision tree model performs best,and itsclassification accuracy rate reached 85.2%.

关 键 词：作文流畅性 作文自动评价 语言学特征 

分 类 号：G424.74[文化科学—课程与教学论]