Article Preview
TopIntroduction
Nowadays, the popularization of online course learning systems, such as Xuetangx (www.canvas.net), makes it possible for students to receive high-level continuing education regardless of time and location limitations. However, given the massive data generated in these online platforms and the lack of face-to-face communication between teachers and students, it is difficult for a teacher to know a student’s learning state. Students have different learning backgrounds and knowledge, which also aggravates this lack of understanding. For example, some teachers believe that students who spend more time watching the teaching videos by the teacher in online courses can master the course well, but that is not always the case, probably due to learners’ lack of required background knowledge in the course subject matter. Moreover, as courses are composed of sets of concepts or knowledge-points, directly modeling the course information will ignore how students learn different concepts and hence cannot consider knowledge-points from the perspective of different students in order to assess the state of their learning. Therefore, the objective of our work is to discover the importance of different knowledge-points to different students and to achieve it by designing a personalized method. Intuitively, if teachers can discover the importance of each knowledge-point to different students, it will help teachers carry out personalized teaching. Based on the learned knowledge-point importance, teachers are able to further suggest other related knowledge to the student to improve her learning efficiency. For example, if teachers know list (a knowledge-point in Python) is important to a student, they may recommend the related cycle knowledge-point. Moreover, the teacher may also need to adjust the teaching strategy for different students, since students have different learning priorities.
In recent decades, applying machine learning and artificial intelligence technology to study student learning processes and improve personalized teaching has always been a research hotspot. However, only a few works are focused on identifying the importance of personalized knowledge-points. For example, Leake et al. (2004) modeled the importance of concepts in concept maps by assessing how a series of potential structural factors combine to affect human judgments of the importance of concepts. Wu et al. (2007) took the value of Hub as the concept importance according to the graph structure characteristics of ontology. They calculated the weight of concept importance by using the iterative method of mutual enhancement of concepts and relations. Wang, Zhang, et al. (2020) constructed a knowledge map of university physics based on a mind map of university physics, in which an expert questionnaire and natural language analysis method were adopted to obtain the importance of knowledge-points. A mind map is an approach to the organization of the human mind that prepares the ground for thinking (Baghestani et al., 2021). Though previous studies represent outstanding achievements, there are still some shortcomings to be addressed. First, in the online learning scenario, such as the Xuetangx learning platform, there is no upper limit on the number of students enrolled in each course, resulting in a large number of students participating in courses. Previous studies are mainly based on expert analysis, are not data-driven, hence inapplicable to the large-scale online scenarios. Secondly, existing methods fail to calculate the importance of every knowledge-point for each student. In addition, as a record of student progress in practice or test, learning-evaluation data (the content of which is shown in Figure 1) provides us a new research opportunity to study the importance of knowledge-points to each student.