基于TextRank和字符级卷积神经网络的小学作文素材自动分类模型研究

随着教育技术与信息技术的融合,实现面向小学生的语文写作自动辅助成为可能。快速自动地进行范文素材的分类入库是实现写作自动辅助的关键。作文素材语义信息丰富、种类较多,若采用现有方法进行自动分类入库操作往往难以取得好的效果。因此,在分析小学作文的类别特征并构建了一个数据集的基础上,提出基于TextRank和字符级卷积神经网络的小学作文自动分类模型。运用基于TextRank的关键句提取模型为范文素材,去除部分冗余的语义信息。应用word embedding对数据集进行文本表示,并将其作为卷积神经网络的输入。通过不断地迭代训练和测试,最终实现了该模型。实验表明了该方法对于作文分类任务能显著地提高分类的性能。

AUTOMATIC CLASSIFICATION MODEL OF COMPOSITION MATERIAL IN PRIMARY SCHOOL BASED ON TEXTRANK AND CHAR-LEVEL CNN

With the integration of education technology and information technology,it is possible to realize the automatic guidance of composition for primary school students.Fast and automatic classification and storage of model materials is the key to achieve automatic guidance of writing.Composition materials are rich in semantic information and various.It is often difficult to achieve good results via normal methods for automatic classification and storage.Therefore,on the basis of analyzing the category features of compositions in primary school and constructing a data set,an automatic classification model for compositions in primary school was proposed based on TextRank and character-level CNN.The key sentence extraction model based on TextRank was adopted to remove some redundant semantic information for the essay materials.The word embedding was applied to express the text of data set and took it as the input of convolutional neural network.The model was realized through continuous iterative training and testing.Experimental results show that this model can obviously improve the performance of composition classification.