基于语义相似度与XGBoost算法的英语作文智能评价框架研究

A study of automated English essay evaluating framework based on semantic similarity and XGBoost algorithm

作文智能评分和评语智能生成能极大减轻评阅专家的工作量、节约人力成本。目前,评分和评语结果的准确性与公平性尚不高。近年来,机器学习和自然语言处理等技术的快速发展,在一定程度上提升了文本分类、机器翻译等任务的性能,但仍有许多新的研究成果尚未应用于作文智能评价。本研究综合了词向量(word2vec)、段落向量(paragraph2vec)、词性向量(pos2vec)和LDA(latent dirichlet allocation)等特征,共同组合为作文的语义表示向量;采用基于kNN(k nearest neighbors)算法的语义相似度模型,得到作文的评语标签;采用基于XGBoost(extreme gradient boosting)的回归模型计算英语作文的评分值;并以900篇大学生英语作文为样本,构造算例进行验证。最后表明,提出的智能评价框架在英语作文自动评分和评语生成的准确性上,都要高于传统方法。

Automated essay scoring and comment generation has greatly released expert human raters from huge workload of evaluating English essays,but up to now,there is still some doubt about the accuracy and fairness of its results.In recent years,with the rapid development of machine learning and natural language processing,etc.,to some extent the performance of text classification,machine translation and the like has been improved.However,quite a number of new research achievements have not been applied to automated essay scoring.This paper presents a semantic representation vector of essays,which is a combination of the features of word2vec,paragraph2vec,pos2vec and LDA(latent Dirichlet allocation);then,the commentary labels of essays are generated through the semantic similarity model based on kNN(k nearest neighbors)algorithm;next,the English essays are scored on the basis of XGBoost(extreme gradient boosting)regression model;finally,900 college students′English essays are taken as samples to verify the results.The case studies show that the evaluating framework in this paper has higher accuracy in automated scoring and comment generation of English essays than traditional methods.