基于监督学习的HSK阅读文本自动分级模型研究

针对HSK（汉语水平考试）各类阅读材料难度判定与等级对应中缺乏有效参照标准和分析工具的问题,以历年HSK真题阅读文本为研究对象,提取文本可读性特征,采用支持向量机、随机森林、极端梯度增强等9种监督学习算法,建立可将自选文本自动归类于相应HSK等级的模型,采用准确率、AUC等多项指标评价各模型的分级效果,并选择最佳模型制成在线工具。结果表明,监督学习在HSK阅读材料文本分析及分级方面具有较高性能,9种模型中极端梯度增强的分级效果最好,准确率为0.913,AUC为0.994。建立的分级模型和在线工具能够以较高的准确率对HSK自选文本进行分级,帮助用户有针对性地遴选文本,提高学习效率。

Research on automatic grading model of HSK reading texts based on supervised learning

Aiming at the problem that there are few effective reference standards and analysis tools available in classifying and grading Hanyu Shuiping Kaoshi(HSK) reading materials, with HSK reading texts in the past years as study object, the text readability features were extracted, and nine supervised learning algorithms, such as support vector machine, decision tree and extreme gradient enhancement, etc., were employed to build a model that could automatically classify self-selected text to the corresponding HSK level. Multiple indicators such as accuracy and AUC were adopted to evaluate the grading effect of each model, and the best model was chosen to design an online tool. The results show that supervised learning has high performance in analyzing and grading HSK reading materials. Among the nine supervised learning models, extreme gradient enhancement is the best, with an accuracy of 0913 and an AUC of 0994. The grading model and online tool can grade HSK self-selected texts with high accuracy, help users select texts pertinently and improve learning efficiency.