Licia
Calvi
University of Antwerp
Department of Romance Languages and Literatures
Universiteitsplein, 1
B-2610 Antwerp
tel. +32 3 820 28 18
fax +32 3 820 28 23
calvi@uia.ua.ac.be
Abstract: In what follows we focus on the notion of self-modifying hyperdocuments as a means to achieve user's adaptivity and we try to motivate it in terms of the user's cognitive processes and learning styles. We show how this has been applied to a successful example of self-modifying hyperdocument, i.e., the on-line course "Hypermedia Structures and Systems" (we refer to [Calvi et al., 97a], and [Calvi et al., 97b] for a detailed description of the work in question).
keywords: self-modifying hyperdocuments, user-adaptivity, dynamic content presentation, dynamic link structure.
If we take the distinction between two different sorts of books, i.e., the books-to-be-read and the books-to-be-consulted, for granted, we should as well identify two different approaches to the paper-based versus Web-based courseware dichotomy. While the books-to-be-read are essentially conceived to be read linearly, from the first to the last page, because the author builds up an argument through the sequence of pages that will bring the reader to a pre-defined conclusion, i.e., the conclusion the author wants the reader to discover, regardless of the nature of the book in question, be it a philosophical essay or a detective story, the books-to-be-consulted are meant to be browsed transversally. There, the reader is immediately pointed to the information she is looking for, skipping from one page to another. This is actually the case of encyclopedias and of textbooks. They are open systems: they contain a finite amount of data but can be used in theoretically infinite ways. And this is precisely where (adaptive) hypertext models succeed, because they provide an adequate framework for free navigation. Books-to-be-consulted are nevertheless also provided with an arbitrary presentation principle, that can be alphabetical for encyclopedias and somehow logical for textbooks. Here, the term "logical" reflects the idea the author has about the reader's expectations and skills. Not to mention the knowledge category the user falls into. If the book is intended to be for beginners, then the author would be inclined to give the content material a structure that matches onto the user's possible knowledge and that foresees an evolving progress from the stage of little or no content knowledge to that of its mastery. So, at this stage, the distinction between textbooks as books-to-be-consulted and books-to-be-read becomes labile. Nevertheless, the system is open, so, once some knowledge has been acquired, i.e., once the author's pre-defined evolution has been achieved, the user is able to use that same textbook as she would do with a normal encyclopedia, i.e., as reference book, e.g. by referring directly to the section about the Positivistic philosophy by skipping all the previous parts concerning the post-Kantians.
But in the last years a new literary poetics has emerged, claiming that all books, both the books-to-be-read as well as the books-to-be-consulted, can be reconceived under the light of the hypertextual paradigm. If this statement implies that, given a finite set of facts, every interpretation, every construction is possible, then its outcome is clearly unsustainable, for how numerous the intersections might be, they can not proceed ad infinitum: one can not expect to derive the demonstration of Fermat's theorem from Joyce's Finnegans Wake, despite the great number of possible interpretations this work is actually open to [Eco, 96]. But if the assumption actually claims that any narration has to reflect, has to realize a specific narrative, what has actually been identified as the hypertextual narrative [Landow, 92], then the above mentioned analogy is true, i.e., all forms of books can be rewritten according to the hypertextual model, and it can even be stretched to the extent that the same entity, namely the textbook, is itself either a book-to-be-read as well as a book-to-consulted. It can indeed self-modify itself in order to fit the user's cognitive skills in terms of learning progress.
This paper presents a framework for self-modifying hyperdocuments which is based on the user's cognitive skills.
A textbook is originally delivered as book-to-be-read in the form of an on-line hypertext. The author provides one only reading order, which the user is forced to follow, under pain of missing completely the message the author is striving to communicate. Here, the reading order is neither intended nor expected to be linear as it is the case for paper-based textbooks, because of the intimate nature of the hypertextual framework. Nevertheless, it is compulsory if the user, supposedly a beginner user, has to evolve towards a stage of content mastery. In practice, this means that propaedeutic material must be processed before more advanced one. And that this more specific information can not be accessed directly from scratch, but only as a consequence of the learning progress. At this point, when the user has acquired some mastery, the textbook becomes no longer suitable as book-to-be-read, but has to transform itself into a book-to-be-consulted, i.e., into a reference book, where the specialized concepts can be referred to immediately without having to undergo the same complete process every time. This can be achieved by updating both the hyperdocument link structure and the content presentation. This would nevertheless be impossible without considering the user's cognitive skills as well as learning styles in information processing.
In a hypermedia system user's adaptivity could be achieved by applying technologies derived from either Intelligent Tutoring Systems (ITS) or Artificial Intelligence (AI) (see [Weber, 96], for an overview of the field). Here, the system generates a model of the user and uses it in combination with some knowledge about the domain of instruction and instructional strategies either to modify the sequence or the quality of information presentation, to support learners and to give users advice, and to define a specific interaction modality depending on the particular user's characteristics, based on the system's current knowledge as well as on the knowledge acquisition process [Weber, 96]. But the same user's adaptivity can be guaranteed more easily by enabling only the link structure and the content presentation mode that match onto the user's present needs, which are assumed a priori as confirming to some pre-defined knowledge stereotypes and to some recognized learning procedure. This conveys the idea that a model of human cognition is preliminary to any application.
Next section shows how this theory has been applied within the framework of the course "Hypermedia Structures and Systems", organized by the Eindhoven University of Technology, which is a successful example of the possibility of self-modifying hyperdocuments (we refer to [Calvi et al., 97a], and [Calvi et al., 97b] for a detailed description of the work in question).
The organization given to on-line documentation mostly depends on the kind of information that has to be delivered. However, whatever the structure may be, the underlying purpose in outlining the subject matter is to make it as simple and as useful as possible [Rojas-Fernandez, 91]. An unstructured or insufficiently structured information space topology can indeed give rise to problems such as a difficulty on the side of the user in understanding the application domain and a consequent navigation breakdown due to uncontrolled linking.
The graph-theoretical methods used to analyze link structures [Botafogo et al., 92], [Begoray, 90] , [DeYoung, 90], [De Vocht, 94] have been able to identify those link structures that are actually unusable, no matter what the content of the nodes is and regardless of the visual aids that may be provided to help the user understand this structure. However, while these methods are able to find link structures which are likely to be unusable, they cannot guarantee that link structures having all suggested values for different metrics will actually belong to highly usable hyperdocuments.
Empirical findings have demonstrated that links are often introduced without a deep consideration of their actual use and usefulness [Campagnoni et al., 89]. Because of the lack of a clear methodology for linking as well as of a common understanding of what the notion of link eventually means, users often happen to be unable to understand why some piece of information is at the end of a certain link and they may also fail to understand why they are sometimes unable to find what they are looking for [Landow, 87]. So, linking one piece of information to another can not only fail to provide the expected benefits, but, in some cases, it even frustrates users. Since the mere presence of links in a hyperdocument seems to suggest that the relationship between the interconnected materials is a significant one, it becomes very difficult for users to ignore them.
So, instead of searching for the "ideal" link structure for a hyperdocument, we opt for a system that adapts the link structure during the learning process by exploiting the user's cognitive abilities. Every user is characterized in terms of a series of episodes, each of which consists of both the information provided in the learning material as well as of the user's own solutions to final tests. The software for the hypermedia course keeps indeed track of all the nodes read by each individual student. In other words, individualization is achieved by storing information from sequences of learning episodes and, on the basis of the specific user's learning history, by triggering the adequate linking structure.
As mentioned earlier, the system does not actually store a user's profile by means of ITS or AI techniques. Conversely, user's adaptivity is achieved by disclosing the fitting link structure and content presentation. In this respect two parameters are of importance:
Whether certain links appear or not depends on the combined knowledge id's a student has acquired. But regardless of the student's knowledge, there are also unconditional links which are always shown.
Using knowledge id's we can make links accessible when the reader is ready for them, and make them inaccessible again after the reader has no need for them anymore. The link structure of the course text therefore changes each time the student follows a link. This technique is used in the courseware for the hypermedia course as follows. When the student first visits the course, the starting page gives some kind of table of contents, but provides no other links than one to a node containing instructions about the use of this courseware. When the student returns from the instructions page, the first three "chapters" of the course text become accessible. These chapters are an introduction, a set of definitions, and a historic overview of hypermedia research and development. The student may follow different paths through these chapters, and depending on the path taken, the link structure within these chapters may also differ. After reading some of the nodes of these chapters, and completing a small (multiple-choice) test at the end of each chapter, links to the more advance chapters appear in the table of contents. While reading, some links will disappear, like, e.g., links to definitions of terms (once those definitions have been read), as well as links to nodes containing unimportant background information (such as the nodes describing the IBM PC, or the Macintosh, which we do not expect students wish to read multiple times). After reading (some of each of) the advanced chapters, a link to the final assignment appears in the table of contents.
From the example of the hypermedia courseware it is clear that a dynamic link structure can be used to eliminate the confusion that could be caused by providing links to information the student is not yet ready for, or that the student has already read and may not wish to find other links to. However, seeing a different selection of outgoing links each time a node is visited can also be confusing. Therefore, the pages with a varying link structure must be limited and well chosen. The table of contents is an excellent candidate for a node with a varying link structure. Through the links in this node large parts of the course text can be enabled at once. In the hypermedia course, most other use of dynamic link structure is in the elimination of links to definition nodes and to unimportant background information.
Similarly, depending on the knowledge acquired by the student, different versions of the content of a node, normally an introductory and an advanced version, are offered. This reflects the idea that, in educational systems, learning normally implies a progression from a situation of unfamiliarity to one of mastery of the knowledge corpus. Such a progression is gradual, episode-based. And it implies as well a consequent shift from the textbook as a book-to-be-read to a book-to-be-consulted.
Acknowledgements: The author is indebted to Prof. P. De Bra for the collaborative research that has lead to several of the ideas discussed in this paper.
[Begoray, 90] Begoray, J.A., "An Introduction to Hypermedia Issues, Systems and Application Areas", Proc. of the ACM Conference on Hypertext, 94--105, 1990.
[Botafogo et al., 92] Botafogo, A.R. and Shneiderman, B.,"Identifying Aggregates in Hypertext Structure", Proceedings of the Third ACM Conference on Hypertext, 63--74, 1992.
[Calvi et al., 97a]
Calvi, L. and De Bra P., "Improving the Usability of Hypertext
Courseware
Through Adaptive Linking", Proceedings of the Flexible Hypertext
Workshop, 1997.
(
http://wwwis.win.tue.nl/~debra/flex97/
)
[Calvi et al., 97b]
Calvi, L. and De Bra, P., "Using Dynamic Hypertext to Create
Multi-Purpose
Textbooks", (to appear) Proceedings of ED-MEDIA'97, 1997.
(
http://wwwis.win.tue.nl/~debra/ed-media97/)
[Campagnoni et al., 89] Campagnoni, F.R. and Ehrlich, K., "Information Retrieval Using a Hypertext-Based Help System", ACM Transactions on Information Systems, 7(3), pp. 271--291, 1989.
[Eco, 96] Eco, U., "From Internet to Gutenberg", lecture presented at the Italian Academy for Advanced Studies in America, 1996. (http://www.italynet.com)
[DeYoung, 90] DeYoung, L., "Linking Considered Harmful. Designing and Reading Hyperdocuments", In Streitz, N., Rizk, A., Andre, J. (Eds.), Hypertext: Concepts, Systems and Applications. Proceedings of the First European Conference on Hypertext, Cambridge University Press, 238--249, 1990.
[Landow, 87] Landow, G.P., "Relationally Encoded Links and the Rhetoric of Hypertext", Proceedings of the ACM Conference on Hypertext, 331--343, 1987.
[Landow, 92] Landow, G.P., Hypertext - The Convergence of Contemporary Critical Theory and Technology. The John Hopkins University Press, 1992.
[Rojas-Fernandez, 91] Rojas-Fernandez, H., "Online and Hypermedia Information Design", Proceedings of IPCC'91, Engineered Communication, IEEE Cat. No. 92CH3087-4, 28--32, 1991.
[De Vocht, 1994] De Vocht, J., Experiments for the Characterization of Hypertext Structures. Masters Thesis, Eindhoven University of Technology, 1994. ( http://wwwis.win.tue.nl/~debra/joep/)
[Weber, 96] Weber, G., "Episodic Learner Modeling", Cognitive Science, 20:2, 195--235, 1996.