Adapting to Prior Knowledge of Learners

Prior knowledge is an important factor that influences the interaction with a hypertext and the learning gain. Our authoring system NetCoach provides a way to assess the users’ prior knowledge and to adapt the course in different ways. We describe how the adaptation mechanism assesses the user’s knowledge with test items, infers the user’s current learning state from this information, and finally adapts accordingly. An evaluation study with an adaptive HTML course demonstrates that this kind of adaptation might reduce the completion time, but retains the learning gain. 1 Online Learning Courses and Prior Knowledge Online learning courses are used by people that differ widely in prior knowledge of the domain. Especially in further education and learning on demand settings some learners will have a background of parts of the course while others are complete beginners. However, regardless of the prior knowledge, everybody should have the same knowledge after course completion. On the one hand users might get bored if they have to work on topics that they are already familiar with. On the other hand they are probably not able to estimate whether they do really know everything on a topic of a course without having seen the chapters. Thus, letting users decide on their own whether they have enough knowledge or not might result in incomplete knowledge acquisition. Moreover, prior knowledge has an impact on the learning gain. When constructing a hypertext authors should considers the users’ prior knowledge [3]. It might be useful to adapt the hypertext’s structure [2] or to provide different advisements [4]. In any case such adaptations require the assessment of prior knowledge. The authoring system NetCoach [6] provides a mechanism both to assess the user’s prior knowledge and to adapt the course accordingly. 2 Adaptation Mechanism NetCoach is designed to enable authors to build adaptive online courses without programming. While authors generate the content by filling in templates and forms, the course functionality including user management, adaptation, communication facilities, and tutoring is provided by NetCoach. Each NetCoach course implements several adaptive features, that require an adequate assessment of the prior knowledge. Based on this information NetCoach infers the user’s knowledge about each chapter and decides how to adapt. 2.1 Assessment of Prior Knowledge Similar to a book, a NetCoach course is structured hierarchically with chapters and subchapters. The knowledge about a subchapter or a set of subchapters is assessed in so called test groups, i.e., a set of test items. Three types of test items are evaluated on the fly automatically: multiple choice tests, forced choice tests, and gap filling tests. Authors may define the difficulty of the test items and a critical value of points that are required for completing the test. After the learner has completed the chapter the items are presented in small groups (e.g., two in a row). Users achieve points for correct answers and loose points for false answers until they reach a critical value. In the beginning of a chapter that has subchapters with test groups a link to a pretest is offered. In this case a random selection of the test groups’ difficult items is presented. In summary, the test items of test groups in subchapters which are usually presented after the chapter are aggregated and now serve as pretest for the superordinated chapter. 2.2 Inference of Current Learning State Each item is evaluated as if it was answered in the test group. The only exception is that the result is weighted (default is 1.5). Afterwards, the current learning state is updated, i.e, NetCoach computes on the fly whether a chapter is solved, inferred, ready or not ready. See Weber, Kuhl and Weibelzahl [6] for a detailed description of the overlay model and the different states of chapters used in NetCoach. Such an inference mechanism requires information about relations between chapters which is stored in a so called knowledge base. NetCoach supports the two concept relations prerequisite and inference [6]. In summary, the inference of a user’s current learning state is done in two steps. First, the items of the pretest are evaluated and points weighted by an factor are assigned to test groups accordingly. Second, if a test group has been solved because the critical value has been reached, this chapter is assumed to be learned and other chapters might be inferred to be learned in correspondence with the knowledge base. 2.3 Adaptation Decision Finally, the learning state is used for two types of adaptations: adaptive annotation and curriculum sequencing. First, links to other chapters are annotated according to the state of this chapter (e.g., a green bullet indicates that the chapter is ready to be visited). Second, NetCoach suggests which page to visit next and thus guides users to chapters that are ready and that are required to complete the course. Chapters that are assumed to be learned are skipped. Thus, the pretests provide an easy way to guide users to those chapters that they should learn. Other adaptation techniques that use the same information would be possible (e.g., link hiding [1] or changing the adaptation method [5]), but are not implemented in NetCoach.