Course Concept Extraction in MOOCs via Embedding-Based Graph Propagation

Proceedings of the The 8th International Joint Conference on Natural Language Processing, pages 875–884, Taipei, Taiwan, November 27 – December 1, 2017 c⃝2017 AFNLP Course Concept Extraction in MOOCs via Embedding-Based Graph Propagation Liangming Pan, Xiaochen Wang, Chengjiang Li, Juanzi Li and Jie Tang Knowledge Engineering Laboratory Department of Computer Science and Technology Tsinghua University, Beijing 100084, China {panlm14, wangxc15, licj17}@mails.tsinghua.edu.cn {lijuanzi, tangjie}@tsinghua.edu.cn Abstract Massive Open Online Courses (MOOC- s), offering a new way to study online, are revolutionizing education. One chal- lenging issue in MOOCs is how to de- sign effective and fine-grained course con- cepts such that students with differen- t backgrounds can grasp the essence of the course. In this paper, we conduc- t a systematic investigation of the problem s.of course concept extraction for MOOC- We propose to learn latent repre- sentations for candidate concepts via an embedding-based method. Moreover, we develop a graph-based propagation algo- rithm to rank the candidate concepts based on the learned representations. We eval- uate the proposed method using different courses from XuetangX and Coursera. Ex- perimental results show that our method significantly outperforms all the alterna- tive methods (+0.013-0.318 in terms of R- precision; p ≪ 0.01, t-test). 1 Introduction In contrast with traditional courses that have lim- ited numbers of students, each online course in a MOOC platform may draw more than 100,000 registrants (Seaton et al., 2014). The students have very diverse backgrounds; and knowledge present- ed in MOOCs is well understood by certain stu- dents, but might be difficult to others. In MOOC- s, we use course concepts to refer to the knowl- edge concepts taught in the course videos, and re- lated topics that help students better understand course videos. Identifying course concepts at a fine level is very important, as students with dif- ferent backgrounds have totally different require- ments. Figure 1 shows an example to illustrate the Figure 1: Example of a video clip and its course concepts problem addressed in this work. It is a clip (about ten minutes long) of video captions from the data structure course in XuetangX 1, one of the largest MOOCs in China. The related course concepts are marked in red and blue. We see that there are a dozen concepts mentioned in this clip. The ma- jor concept is “quick sort” (marked in red), but the content is related to many other prerequisite concepts such as “divide-and-conquer”, “bubble sort”, and “unstable sorting algorithm” (marked in blue). For students with a computer science back- ground, those concepts would be easy; however, for students from other discipline areas, the con- cepts may be completely new. With the identified course concepts, we can build an interactive learn- ing environment to help individual students better grasp the knowledge in a course. For example, for a non-science student, the system may display the definition of “unstable sorting” and “uniform dis- tribution”; while for a science student, the system could recommend advanced concepts or potential applications of “quick sort”. Course concepts were previously …

Content truncated to 500 words