Abstract: The WWW provides a unlimited resource of materials for learning. However, learners have to find their way through links which sometimes are not relevant to their goals. We have developed a tool for authoring of adaptive CAL courses, called "Dynamic Courseware Generator" (DCG). It generates individual courses according to the learnersí goals and previous knowledge and dynamically adapts the course according to the learnersí success in acquiring knowledge. The DCG runs on a WWW server. The learner is a client; s/he receives a course corresponding to his/her learning goal and is guided through a set of teaching materials on the WWW.
With the emergence of the WWW it became possible to provide learners with unlimited access to teaching materials. Authoring became easy since to produce such materials (html-files) one can already use common text-editors. However, learners in the WWW face one significant problem: they have to cope with a huge amount of materials and to navigate through links which sometimes are not relevant to the learnerís goal. Often they get lost in the hyperspace, forget where they started and loose a focused perspective on the field.
CAL provides one simple solution to the problem by limiting the browsing activity of the learner to a pre-defined sequence of teaching materials (course). This solution is, however, quite inflexible to accommodate various learnerís goals and success in acquiring the material. This problem can be solved by providing learners (clients) with access to a tool for automatic generation of adaptive courses on a server
We have developed a tool, called "Dynamic Courseware Generator" (DCG), for automatic generation of individualized courses according to the learnerís goal and previous knowledge which can dynamically adapt the course according to the learnerís success in acquiring ,,. It has been tested in several domains: the structure and functioning of electrical devices, medical diagnosis and arts (jazz). The DCG allows:
We have recently implemented the DCG on a WWW-server and now it can be used for domain-authoring and automatic generation of adaptive courses on the WWW.
The main idea of the DCG architecture is the explicit representation of the concept structure of the domain, separated from the teaching materials which the student sees on the screen. The concept structure is used by the system as a road-map to generate a plan of the course. Given a certain goal-concept that the learner wants to acquire and a student model containing the concepts already known by the student (initialized with a pre-test), the planner searches for sub-graphs that connect the concepts known by the learner with the goal-concept. One of these plans is selected and offered to the learner to follow. The learner sees a sequence of teaching materials related to each concept from the plan. At every point the learner can choose to be tested on his/her knowledge of the current concept and s/he will be given a set of test-items. A Bayesian Student Model is created for every learner which is an overlay over the concept structure. Updating of the knowledge probability of a given concept takes place when the learner solves a test-item. The updating coefficient depends on the itemís difficulty and on the correctness of the answer. If the learner is not able to achieve higher than a given threshold score for a given concept, that is needed to proceed further with the plan, a new planning takes place.
We adopted a client-server architecture for the system on the WWW (see Figure 1). The DCG is placed on a WWW-server; it offers teaching and authoring services. A client can be every WWW-browser. Learners and authors /teachers can be clients and use the teaching and authoring services offered by the DCG-server. The teaching materials used in the courses generated by the system are also distributed on various sites in the WWW.
Figure 1: Client-Server Architecture of the DCG on the WWW.
The authors can also be distributed over the net on their sites (see Figure 2). Authoring consists of two parts:
A special graphical editor for concept structures allows the creation of concepts, inter-relating them with various types of semantic relations (e.g. abstraction, aggregation, analogy, temporal, causal, etc.). One-directional 1:n and n:1 relations are used to represent hierarchies with respect to abstraction, aggregation and causal relations, and bi-directional 1:1 relations to represent analogical and temporal relations.
Every concept and relation can be linked to the http-addresses of one or more teaching materials (represented as html files) and testing materials (html-files with an attached JAVA procedure which carries our the interaction and answer evaluation). One teaching material can address one or more concepts/relations and one concept/relation can be presented with one or more teaching materials. For the creation of the teaching and testing materials the author can use every html editor or reuse existing html sources. The creation of testing materials requires only a link to the JAVA procedure and assigning the right answer to the question.
Figure 2: Authoring
Figure 3: Learning with the DCG.
The learner sends a request for a course to the DCG-site stating the desired domain and learning goal. The DCG sends them a preliminary test to initialize a student model for each learner and enters him/her as a user. Then the learner receives an individual course-plan and a copy of a JAVA program, called "Executor" which runs on the learnerís site. The Executor tests the studentís knowledge on the concepts and updates the Student Model locally at the learnerís site. In case of studentís inability to perform successfully on the test related to a certain concept that is needed for continuing with the plan, the Executor contacts the DCG site; it sends the Student Model there and requests from the Planner a new course-plan, appropriate for the new state of knowledge of the student. The learning process is presented in Figure 3.
The DCG on the WWW has been experimentally implemented on three local servers at the Federal Armed Forces University with domain covering the material of the first two lectures of a course on Computer Based Learning Systems (CBL) - introduction to the field, history and technologies applied in CBL (see Figure 4). The goal of the evaluation of the DCG on WWW was to see how such a system can be integrated in university teaching. DCG in WWW was experimentally used by twelve students taking the course of CBL. Our experience showed that the DCG on WWW can be easily and usefully integrated in the existing organization. The following main applications were outlined:
We believe that ITS-authoring tools with distributed architectures like the DCG can provide a good start towards WWW-based adaptive educational systems.
Figure 4. The Interface of the DCG
The DCG on the WWW is an example of a ITS Authoring System scaled up to work on the WWW. Another example of such systems is ELM-ART . They are similar in that they use a secondary concept-based indexing of the educational materials which allows building a secondary navigation map. However, there are significant differences between ELM-ART and DCG, for example, in the level of adaptivity provided by ELM-ART and the DCG. In ELM-ART the adaptation is only in the presentations of individual concepts while in the DCG also the navigation through the concept structure is tailored to the individual student. The format of the teaching materials used in both systems is different. In ELM-ART the presentations are automatically generated by the systemís knowledge base and the "special program". LISP is needed to ensure the automatic frame-based representation of the annotated html-documents as well as to generate the adaptive presentations at run time. The DCG can use as a teaching materials any html-file. There is also a difference in the style of teaching, provided by ELM-ART and DCG. ELM-ART provides a new hyperspace for the student to explore, which is in line with the constructuvist tendencies in learning environment design. The DCG provides goal-oriented courses in the traditional CAI sense. Some other adaptive teaching system on the WWW exist , , which are similar to ELM-ART, but differ significantly from the DCG.
We have developed a tool for authoring of adaptive CAL courses, called "Dynamic Courseware Generator" (DCG). It generates individual courses according to the learnersí goals and previous knowledge and dynamically adapts the course according to the learnersí success in acquiring knowledge. The DCG runs on a WWW server. The student receives from this server an individualized course targeted to a specified by the learner goal. Afterwards, s/he is adaptively guided by the course through a space of teaching materials on the WWW. Unlike other CAL courses and educational hypermedia on the WWW, a course produced by the DCG is interactive, it tests the learnerís knowledge and dynamically adapt the course-plan.
 Brusilovsky, P., Schwarz, E., Weber, G. ELM-ART: An Intelligent Tutoring System on World Wide Web, in Proceedings of ITSí96, Lecture Notes in Computer Science No 1086, 1996, Springer: Berlin.
 Diessel Th., Lehmann A., Vassileva J. Individualized Course Generation: A Marriage Between CAL and ICAL. Computers and Educ., Vol. 22, No.1/2, 1994, 57-64.
 Ibrahim B. & Franklin, S. Advanced Educational Uses of the WWW. Proceedings of the Third WWW Conference, 1995.
 Kay J.& Kummerfeld B. An Individualized Course for the C Programming Language. In:Proceedings of the Second WWW Conference, 1994, Chicago, Elsevier.
 Vassileva J. Dynamic Courseware Generation within an ITS-shell Architecture. In: Proceedings of ICCAL'92, International Conference on Computer Assisted Learning,Lecture Notes in Computer Science No 602, 1992, Springer: Berlin-Heidelberg, 581-591.
 Vassileva J. Dynamic Courseware Generation: at the Cross Point of CAL, ITS and Authoring. In: Proceedings of ICCE'95 - International Conference on Computers in Education, Singapore, 5-8 December, 1995, 290-297.