基于神经网络的英语作文自动评分模型研究与实现-计算机应用技术专业论文.docxVIP

万方数据 万方数据 摘要 摘 要 英语作文是大规模英语考试中的一种必备题型，无论是在我国的高考、研究生考 试还是在国外的 TOEFL、GRE、IELTS 中，英语作文都是检测英语学习者语言综合 运用能力的一项重要指标。目前，国外已经开发出了一些针对母语为英语的作文自动 评分系统，但是针对中国大学生英语写作特点的作文自动评分研究仍有待加强，因此 针对国内大学生的作文自动评分研究是一项具有实际意义的研究课题。 本文通过对作文评分的标准、影响作文评分的因素、特征提取方法、人工神经网 络、自然语言处理等相关工作的研究，提出了一种基于 BP 神经网络的作文自动评分 模型，运用自然语言处理技术对文本集进行词性标注和句法分析，统计各项文本特征， 然后采用特征提取方法对特征进行筛选和分组，分别提取作文的词法、句法、结构、 内容特征，并利用上述特征构建出了四个不同的神经网络拟合各项特征与作文评分之 间的映射关系，并计算出作文的词法评分、句法评分、结构组织评分和语义评分。通 过与英语教师讨论和多次实验确定词法评分、句法评分、结构组织评分和语义评分在 作文评分过程中所占有的比重，最终得到作文的综合评分。 最后对本文提出的作文自动评分模型进行了测试，实验表明本文提出的模型能够 对学生单词和句子的掌握和运用能力、文章组织结构以及语义内容进行有效的评估。 关键词：英语作文自动评分；神经网络；词性标注；句法分析；语义分析 I Abstract Abstract Abstract English composition is an essential part for large-scale language tests, such as domestic entrance for college, Graduate Entrance Examination or foreign TOEFL, GRE, and IELTS. It is a key detection index to reflect the comprehensive language proficiency for English learners. At present, there have been some automatic scoring systems from abroad aiming at Anglophone, but the research for Chinese college students has yet to be further improved. Hence, the research aiming at domestic university students is still a meaningful topic. This dissertation proposed an automatic scoring model based on BP neural network through the study of essay scoring criteria, factors affecting marking compositions, methods of feature extraction, artificial neural networks, natural language processing and other related research. Firstly, natural language processing technic is used for POS tagging and syntax analysis on text set, in order to collect various features that can reflect text language quality and organizational structure. Next, features extraction method is used to filtrate and group features, followed by the extraction of the features such as lexical, syntactic, organizational structure. Four corresponding neural networks are constructed respectively to fit the nonlinear relationship between the group of features and composition scoring. Finally, the pr