习题的关联分析及其向量化表示方法

郭娜,路梅,赵向军

GUO Na，LU Mei，ZHAO Xiang-jun

（School of Computer Science and Technology,Jiangsu Normal University,Xuzhou 221116,China）

Received:2015-11-26 Revised:2016-05-12 Online:2017-10-25 Published:2017-10-25

摘要：

随着互联网+教育的深度融合以及移动终端上电子习题的推广使用，学生的学习过程数据可以被实时获取，充分利用这些过程

数据，及时定位学生的知识病灶，开具有针对性的辅导处方，实现知识的按需推送，对于减轻学生的简单重复劳动，提高学

习效率将会产生积极影响。试图通过分析在线习题系统的答题数据,发现学生的知识掌握规律，根据错题的伴生状况捕获习题

的相关性。为此，构建了题向量化模型，提出了题向量表示的新方法，设计了负采样训练算法，并用程序实现了上述算法。

经过实际在线系统的相关数据训练，获得了相应题向量，而后利用题向量的向量运算，可方便查找相同习题、相同知识点习

题以及相近知识点习题等，可根据学生错题个案，推断其知识掌握的其他薄弱环节。

关键词: 题向量, 题向量化模型, 知识推动, 教育智能

Abstract:

With the extensive application of Internet+ in education and the popularity of electronic exercises on mobile

devices, the learning data of students can be captured in real time. Mining the learning process data can

locate the students' weak points of knowledge, carry out targeted counseling prescriptions and push on-demand

knowledge, which has a positive impact on students in terms of reducing simple repetition and boosting

learning efficiency. We analyze the data of students' answer to the online exercises and propose a novel

vector representation of exercise. The correlation between exercises is captured by analyzing the data of

wrong answers to the exercise by different students and the exercise-to-vector (xcise2vec) model is

constructed. We also design a negative samples training algorithm and implement it in Linux. The exercise

vector can be obtained by training the data generated from the online system. Then, we can search for

similar, equivalent and interrelated exercises via exercise vector operation, and deduce other knowledge weak

links from the status of his answers. Experimental results verify the effectiveness of the proposed method.

Key words: exercise vector, exercise-to-vector(xcise2vec)model, on-demand knowledge, education intelligence

郭娜,路梅,赵向军. 习题的关联分析及其向量化表示方法[J]. 计算机工程与科学.

GUO Na，LU Mei，ZHAO Xiang-jun. Correlation analysis of exercise and its vectorization method[J]. Computer Engineering & Science.

PDF