基于Bloom分类法的CS1试题数据集的构建及其自动分类  

Construction and Automatic Classification of CS1 Test Questions Dataset Based on Bloom's Taxonomy

作　　者：董荣胜[1] 卫晨雨 胡杰 乔宇澄 李凤英[1] DONG Rongsheng;WEI Chenyu;HU Jie;QIAO Yucheng;LI Fengying(Guangxi Key Laboratory of Trusted Software,Guilin University of Electronic Technology,Guilin,Guangxi 541004,China)

机构地区：[1]桂林电子科技大学广西可信软件重点实验室,广西桂林541004

出　　处：《计算机科学》2023年第6期175-182,共8页Computer Science

基　　金：国家自然科学基金(62062029)。

摘　　要：课程评估是教学改革的一个关键环节,涉及教学案例、试题以及课堂教学等方面的内容。针对计算课程的试题评估,引入Bloom分类法,以普林斯顿大学和桂林电子科技大学“计算机科学导论”课程(CS1)的试题为语料库,给出针对CS1的Bloom分类法认知过程维度和知识维度的相应动词种子库和名词种子库,对试题所能达到的Bloom分类法二维矩阵的位置进行标注,构建CS1试题分类数据集。采用机器学习技术,给出CS1试题自动分类模型TFERNIE-LR,该模型由CSTFPOS-IDF算法、ERNIE模型和LR分类器3部分组成。CSTFPOS-IDF算法是在TFPOS-IDF算法的基础上,通过计算课程关键词权重因子,来提高模型对计算课程关键词的关注程度,生成词权重。同时,基于实体知识增强预训练模型ERNIE进行试题词语级向量嵌入,组合词权重和词语级向量生成用于自动分类的试题文本向量。最后,采用LR分类器将试题自动分类到Bloom分类法二维矩阵。实验结果表明,TFERNIE-LR模型具有良好的性能,在认知过程维度和知识维度上的加权精确率分别达到了83.3%和96.1%。Curriculum evaluation is a key link of teaching reform,which involves the evaluation of teaching cases,test questions and classroom teaching.In order to evaluate the test questions of computing courses,this paper introduces Bloom's taxonomy,and takes the test questions of“Introduction to Computer Science”course(CS1)of Princeton University and Guilin University of Electronic Science and Technology as corpus,and the corresponding verb seed bank and noun seed bank for the cognitive process dimension and knowledge dimension of Bloom's taxonomy for CS1 are given,the positions of the two-dimensional matrix of Bloom's taxonomy that could be reached by the test questions are manually labeled,classification dataset for CS1 test questions is constructed.Machine learning technology is used,the automatic classification model TFERNIE-LR of CS1 test questions is given,which is composed of CSTFPOS-IDF algorithm,ERNIE model and LR classifier.CSTFPOS-IDF algorithm is based on TFPOS-IDF algorithm,by the weight factor of the keywords in computing discipline,CSTFPOS-IDF algorithm pays more attention to the keywords improves and generates the weight of words.At the same time,the entity knowledge enhanced pre-training model ERNIE is used to embed the word level vector of test questions,and the combined word weight and word level vector are used to generate the text vector of test questions for automatic classification.Finally,the LR classifier is used to automatically classify test questions into Bloom's taxonomy two-dimensional matrix.Experimental results show that the proposed TFERNIE-LR model has good performance,and weighted-P in the cognitive process dimension and knowledge dimension reaches 83.3%and 96.1%respectively.

关 键 词：Bloom分类法 课程评估 CS1试题分类数据集 动词种子库 名词种子库 自动分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]