Research on the Memory Processing Rules of Children in Tang Poetry Reading Using Internal Feature Extraction and Deep Neural Network Method

There are differences in children's cognitive abilities towards Chinese Tang poetry. Several Tang poems are easier for children to learn, while others are difficult. Thus, it is necessary to understand what characteristics of Tang poetry make children find difficult to learn. This problem was modeled as a binary classification problem in machine learning. Firstly, a Tang poetry recitation test was conducted on children, and the average score of all children on each poem was calculated. After data processing, a measure of the difficulty of Tang poetry was determined as the dependent variable for the study. Tang poetry has many characteristics, such as theme type, publication period, and style depending on whether it rhymes, the number of words, number of sentences, sentence length, and whether there are rare characters. These features were selected as independent variables. We constructed two classification models based on the classification and regression tree (CART) and deep neural network methods. The F1 scores of the methods were 80.3 and 84.8%. Although the accuracy of the CART model was slightly lower, it still showed strong interpretability for the characteristics of Tang poetry that were difficult for children to learn. The results can assist in practical teaching.