Research on TextCNN-based Evaluation of Rationality of Narrative Text Structure

Abstract

Automatic Essay Scoring refers to the use of computers to score composition by some technologies. This process does not require human intervention. Rational text structure analysis is an important part of automatic essay scoring. However, the study of text structure is still in its infancy, ignoring its importance to the evaluation. Existing research lacks a corpus for the evaluation of text structure. The recognition of text components mostly uses artificial experience for feature selection, and evaluation model is established based on them. To figure out these problems, this paper refers to the curriculum standards, works with experts to build text structure standard and labeling method, and formulate corresponding labeling specifications. Finally build a corpus of a certain scale. TextCNN are used to build a model for the text structure. Model treats each article as a whole for training, and realizes the use of deep learning algorithms to make the model automatically evaluate. The results in test set show that in the constructed narrative composition corpus for grades 5-9, the accuracy of the model can reach 72.4%.