Article Preview
Top1. Introduction
In recent years, with the popularization of Internet technology, major changes have taken place in the education field (Martin & Ndoye, 2016). Online learning has broken through the limitations of time and space, making knowledge acquisition more flexible. More users are no longer relying solely on offline physical classroom learning. Online education has solved the problem of sharing educational resources and greatly optimized the allocation of educational resources (Dietz-Uhler B& Hurn, 2013).
Adaptive learning system fully takes advantage of machine learning and data mining, and has a good curriculum of self-adaptive recommendation, providing adaptive learning content and learning paths for different learners. Therefore, it has received extensive attention from the education community and the computer industry. Peter Brusilovsky first proposed the concept of adaptive learning in 1996 (Brusilovsky, Eklund J & Schwarz, 1996) and proposed a general-purpose model for an adaptive learning system. The picture in Figure 1 presents the general model of an adaptive learning system. Overseas has also developed InterBook, ELM-ART (Brusilovsky, Schwarz, Weber & 2001), iWeaver (Wolf C, 2007), AHA! (De Bra,2007) and other adaptive learning systems.
In the universal model, the student model is used to represent the learner's current knowledge status, cognitive level, and hobbies. Through summarizing and analyzing the existing student models (Peter, Sibel& Julio Guerra, 2015), it provides an important basis for the recommendation of the engine based on the adaptive learning system recommendation. The accuracy of the statistics and analysis of the data directly affects the quality of the system recommendation service. However, in the previous studies, most of the student model construction process mainly considers the learner's basic information. The study of the learner's cognitive level often uses the static test database or the traditional Computer Adaptive Testing (CAT), and cannot use the test database dynamically (Alrifai, Gennari &Tifrea, 2012).
Figure 1. The general model of an adaptive learning system
However, candidates with the same or similar scores (capabilities) may also have different cognitive states and different knowledge processing processes. Cognitive Diagnosis (CD) was therefore applied.
The difference between cognitive diagnosis theory and traditional CAT is that it does not provide a single score as the subject's overall assessment (Lee, 2013), but rather pays more attention to the subject's individual condition - its knowledge structure. The current research on cognitive diagnosis theory focuses more on the learning and improvement of algorithms and the basic research of theories, and it is less applied to the specific process of educational learning (Didik, 2014).
Therefore, this paper proposes a CAT hybrid model based on cognitive diagnosis theory(H-CD). According to the actual situation of the learner, the fixed question bank and the question bank exist simultaneously, and has a dynamic selection strategy, which is based on the learner's different knowledge structure. The test database dynamically updates the student model, providing an effective guarantee for the adaptive system dynamic recommendation.