Cognitive models for structuring hypermedia and implications for learning from the world-wide web

John Eklund, Faculty of Education, University of Sydney 2006
Keywords: hypertext, hypermedia, learning


Hypermedia Systems (HMS) and the derivative Multimedia Systems (MMS) are based on Hypertext, which is a non-sequential, non-linear method for organising and displaying information in the form of text, graphics, animation, sound and video. HMS are based on the assumption that the student's interpretation of the courseware is more meaningful than that of the expert or the author. They rely on interface design and the provision of advanced navigational tools such as concept maps or graphical browsers, and levels of search indicators to orient the student and to allow the learner to make an informed decision about where next to proceed in the tutorial.

Hypertext (Nelson, 1978) defines associative links between nodes, which are one or many screens of information. The node-link structure in the knowledge base allows the user to move through an information space utilising navigational tools. There is, inherent in implementations of hypermedia courseware, an author/expert defined structure of the material, simply in the sequencing of the nodes and the availability of associative links between them, but it is at the user's discretion that this sequence is followed. This has been the predominant approach to structuring hypertext-based courseware to date. It is based on the idea that an expert's sequencing and linking of nodes, combined with a domain referenced design of interface ergonomics, provides a knowledge structure which reflects the way learning typically takes place in the content area. However, a fundamental limitation of this approach is that not all students are typical. The sequencing of the material and the links are fixed, and not dependant on the individual user's responses or actions. Unlike intelligent systems, hypertext-based systems are most often a static, non-adaptive learning medium. They do not teach, but instead provide the student with an excellent opportunity to learn of their own accord, and have been described as a non-pedagogical technology (Duchastel, 1992), as they possess minimal structural knowledge of content.

While intelligent systems have been criticised for their structured tutoring and their embedded assumption of the expert's model of domain knowledge as the basis for teaching and remediation, HMS have suffered from a lack of structure or expert guidance in the instructional sequence. A common problem for novice users of such systems is one of disorientation, or "getting lost in hyperspace", and an established research area is the study of student paths or 'trails' through the material (Misanchuck & Schweir, 1992; Santago & Okey, 1992). This empirical approach allows researchers to look for relationships between learning outcomes and navigational paths, with the view that the characteristics of learner paths for high achievers provides guidelines for maximising learning through interface design. Consideration of interface ergonomics is a means of addressing navigational problems users of hypermedia experience, and a way of avoiding representing domain and user knowledge in hypermedia. This approach is common in the literature yet has not yielded significant gains in addressing navigational problems. One possible explanation for this is that well grounded techniques and principles of instructional design have successfully been applied to interface design, and there may be limited scope for improvement given current technologies.

Hypermedia provides advanced navigational tools

One way of alleviating the problem of users becoming lost is through the provision of a variety of navigational tools. These tools available in HMS are categorised by de La Passardiere & Dufresne (1992) as punctual, structural or historical. Punctual aids include buttons which offer transport to some other place in hyperspace (usually with some indication of the destination), as well as help buttons. Structural aids provide users with alternate perspectives on their position and include overview maps, local maps, fisheyes, filters and indexes. Overview maps allow the user to 'zoom out' to see nodes and links, while local maps show links to the current node only. Fisheyes are devices for 'zooming in' on a node. Filters reduce the complexity of displaying links between nodes in an overview or local map, and indexes are structural tools which provide a way of organising the hyperspace into a hierarchy. Historical aids attempt to place users in the information space by showing them where they have been. They include history trails, footprints, and landmarks. History trails allow the user to review the trail through the material, without locating that path using other tools. Footprints are system generated marks showing when a user has passed through a node, while landmarks are user generated for the same purpose.

While the navigational features of hypermedia provide opportunities for students to undertake a non-linear use of the material, that is one which best suits their notions of where to move within the courseware, some studies (eg Messing, 1990) have shown that students tend to adopt a linear pattern of review similar to that taken with a book. Further studies (eg Santiago & Okey, 1992) have attempted to classify learners according to their beliefs about whether they have the ultimate responsibility for navigation of an environment. Describing these learners as 'internal', (accepting that control is contingent on their actions and knowledge), or 'external', (attributing their success in the environment as a function of external variables outside their influence), has provided researchers with a rationale for the student's navigational behaviour in that domain. Other learner centred factors include the influence of prior knowledge, both specifically in the domain under investigation as well as with HMS in general. Not suprisingly it has been found that a knowledge of the subject matter correlates highly with an ability to navigate in a non-linear way through the information space (Ohlsson, 1992); and that a knowledge of hypermedia environments in general also predicts a greater use of learner initiative in the use of available navigational tools (Stephenson, 1992).

Adaptive hypermedia

Considerable research effort is also being directed toward applying the understanding gained from empirical studies of student traits using computer interfaces to create adaptive systems with an interface management approach (eg Pitz, 1994). In these systems, the interface is altered on the basis of several stereotypical user categories. The systems attempt to trace student knowledge and provide individual advice. For example, Yazici, Muthuswamy & Vila (1994) used an intelligent system approach for interface management in decision support systems. They constructed an Expert System Interface Manager (ESIM) which contained a knowledge base of user cognitive characteristics, display preferences, decision tasks as well as the rules to select appropriate display formats. The student model is used to dynamically alter the display characteristics of the interface to suit the needs of the learner.

Other adaptive hypertext-based systems include that by Carlson & Gonzalez (1993), who used hypertext as a basis for representing knowledge in the conceptual design and prototype implementation of a set of knowledge templates and applied this design to writing. The knowledge structures used in this design strongly reflect cognitive models on which the writing process is based. The semantic net structure used was constructed with a cognitive process model embedded in the system's architecture. This was achieved through seven levels where the unstructured text is progressively organised into the finished product.

Work is ongoing at the Swedish Institute of Computer Science (SICS) and Stockholm University (Karlgren et al, 1994), where a national project is underway developing adaptive hypermedia. This work uses adaptive search and filtering mechanisms mainly based on user stereotypes. The stereotypical knowledge is derived from observations of several users, and utilises clustering and direct user input. This method avoids the need for very advanced organisation of hypermedial knowledge.

In a paper by Vassileva (1994) a practical architecture was proposed for user modeling in a hypermedia-based information system as a means of controlling a novice's poor use of hypermedia links. This work recognises the navigational problems of novice users and applies a knowledge structure to the hypertext as a network of goals in a hierarchy. This control was achieved by altering the size of the browsing space on the basis of the user's knowledge. This adaptive, self-improving architecture was implemented for an information system for hospitals. This is a task-based domain which assumes a fixed set of goals, and the hypermedia is organised as a network of hypertext links, goal-topic links and goal-hierarchy links. This adaptive browsing support uses a user-model based on empirical analysis of tasks performed by users.

Kobsa, Müller & Nill (1994) developed an adaptive user modeling system based on hypertext to address comprehension problems experienced by users. The design of this system recognises the two major problems users of HMS face - navigation and comprehension. In this implementation, hypertext objects alter on the basis of the user's current knowledge state. The system's knowledge representation is based on the assumption that users who deselect or bypass information do so because they are familiar with it, and users who request information or follow a path are motivated by a desire to acquire the knowledge.

Mathe & Chen (1994) have developed a user-centred approach to adaptive hypertext which is based on the relevance of the information to the user. The user is asked to indicate, throughout the dialogue, an assessment of the 'interestingness' of a topic and this determines a user-profile. The information relevance model does not require a highly structured hypermedia as it does not reflect the expert's appraisal of the student but the student's own opinion of the relevance of the material. The system adaptively advises the student through the information space on the basis of the student's ongoing indication of what is personally relevant to them. This is a domain-dependant solution to structuring knowledge in hypertext which relies on student input and avoids stereotyping. It does, however, rely heavily on the user's own assessment of their needs, which may not always be correct. For student's to effectively be able to select a sequence and strategy according to their needs, they need to have a substantial knowledge of content in the first place.

Designing computer-based environments for learning

Brown (1990) and other writers (Anderson, 1992; Corbett & Anderson, 1990, Chan et al, 1993, Eklund, 1993) have explained how current theories and architectures in intelligent systems have impacted upon learning theories, and vice-versa. Corbett and Anderson's (1992) description of the LISP Intelligent Tutoring System, namely LISPITS, which instructs students in the programming language of LISP, is based on theoretical principles derived from a theory of cognition proposed by Anderson known as ACT* (Anderson, 1983, 1986). ACT* theory proposes that human problem solving is enabled by a set of production rules, and can be turned into a formal set of well-ordered rules about instruction. Briefly, this is that behaviour is goal-driven; that there is a distinction between declarative and procedural knowledge; that the acquisition of procedural knowledge occurs through a sequence of productions of a declarative nature, which through repetition and practice are collated into larger productions; and finally that in the early stages of learning a working memory is needed before productions are built. LISPITS is a well-studied intelligent tutor founded on the ACT* principles, and has been criticised for its authoritarian teaching strategies and behaviourist approach to learning.

VanLehn (1992) notes that there are tremendous difficulties in attempting to base a wide variety of courseware from varying knowledge domains on any one model of human cognition, and points to domain-specific theories as a means of understanding human learning. It has been the tradition in intelligent systems research to instantiate a theory of learning through a program, and an evaluation of that program is often viewed as reflecting that of the model of cognition on which it is based. The implementation is an artefact of a specific theory and its evaluation reflects the validity or otherwise of the theory.

Cognitive models for structuring hypermedia

Jonassen (1991) has outlined issues related to integrating knowledge structures within hypertext systems as similar to those of integrating knowledge acquired from HMS into the learner's own understanding. Structuring hypertext for effective learning is thus based on structures of learning, or cognitive models, within the learner. This paper defines domain knowledge as divided into three components: knowledge of declarative elements, of procedures, and of structures. Declarative knowledge elements are isolated components that rely on a working memory (eg I know a2 x a3 = a5 but I don't necessarily know why). Structural knowledge is knowledge of the relationships between elements of declarative knowledge (eg I know a2xa3=a5 and this is because a2=axa and a3=axaxa and thus a2xa3=(axa)x(axaxa)=a5). Procedural knowledge is the ability to effectively group a number of these declarative elements and their links into a coarser grained "procedure" and apply it without needing to recall the contents of the procedure in detail, (eg I just use a2xa3=a5 automatically).

Pedagogical knowledge also contains a knowledge of the structure of the domain, that is the interrelationships between concepts and instances; as well as knowledge of the state of the student's understanding of content at any time; and knowledge of a variety of tutorial strategies and a means for selecting a strategy that will best facilitate learning according to these first two states.

HMS exemplify constructivist approaches to learning (see Jonassen, 1991; Cobb et al, 1992), where learning is regarded as the formation of mental models or "constructs" of understanding by the learner. In this view of learning, the students actively build knowledge based on previous understanding by dynamically interacting with the learning media. One practical outcome of this theory of human learning is that the learning medium must create the situation where the learner has the freedom to exercise judgement about what is to be learned and at what pace. HMS are a suitable delivery mechanism for courseware which embodies constructivist approaches as they are structured to allow the learner to take more control of lessons. The success of the learning relies on the interest, intelligence and ability of the learner to make decisions about lesson sequence, timing and emphasis.

It has been suggested (Jonassen, 1992) that HMS are particularly useful in facilitating learning because hypertext structures reflect a model of learning based on schemas. In schema theory, learning is the accumulation and organisation of knowledge structures. These knowledge structures are a representation of the organisation of ideas in our semantic memory. Each knowledge structure exists as an object, idea or event as well as a set of attributes which link it to other knowledge structures. As we learn, we gain new structures and links, adding information to existing structures (also known as accretion), or alter existing structures through a process of restructuring. Restructuring also involves grouping knowledge structures into procedures, or coarser-grained schemas. Our knowledge exists in a semantic memory which is a network of interrelated concepts.

In the semantic net architecture, the knowledge in the courseware is organised as concept nodes, which in Hypercard(TM) or Toolbookreg. may be one or more likely a number of cards, or in Authorware(TM) a group containing a sequence of presentation and interaction icons. Nodes are connected via navigational links under buttons and menus. The nature of the button or menu describes the association between nodes.

Figure 1: A Semantic network for Hypercard

A similar model of learning stages proposed by Chan et al (1993) called OCTR suggests four stages in human learning: orientation (relating prior knowledge), coaching (apprenticeship learning), tuning and routinization (practice with gradually more student autonomy). This model emphasises the constructivist view and also assumes knowledge consists of units and links. The distinction is made between strong and weak links. The four stages in learning involve the creation of weak and strong links between "proper old knowledge" (p. 259) and new material. The stages in the learning model are qualitatively explained in cognitive terms through processes of connection (weak links are created between old knowledge and new knowledge), accretion (knowledge is expanded with many new weak links created), articulation (links are strengthened while some are deleted), and solidification (units and links are strengthened). Chan et al demonstrate the links in the design of a computer system through a tree diagram of the knowledge architecture, represented as a learning goal hierarchy. Each node in the hierarchy possesses a "cluster of episodes" (p. 262) which is essentially a small unit of related material from the domain.

Figure 2: From Chan et al (1993) p.257

Conclusion and implications for learning from the World-Wide Web

Mosaic and Netscape are based on hypertext and rely on the user's ability to make informed decisions about where next to browse. They underlie a constructivist approach where the learner accepts control of the learning. If learning is facilitated by the formation of node-link structures in the learner, then this needs to be reflected in the construction of the knowledge base according to the way learning takes place in the domain.

We have discussed the cognitive models suggested by schema theory and OCTR, and in these theories, learning is the accumulation and organisation of knowledge structures, and are represented as nodes and links between them. The nodes are declarative elements of knowledge while the links represent procedural and structural understanding. In ACT* theory, the acquisition of procedural knowledge occurs through a sequence of productions of a declarative nature, which through practice are collated into larger productions. In other words, through repetition a node-link becomes simply a node, an element of declarative knowledge, and the granularity of the knowledge in the learner alters.

These considerations lead to a number of possibilities to maximise learning in hypermedia environments in general, and from the Web in particular:


Anderson J (1983) The architecture of cognition. Cambridge, MA: Harvard.

Anderson J (1992) Intelligent tutoring and high school mathematics. In C Frasson, G Gauthier, G McCalla (eds) Intelligent tutoring systems: Second intenational conference, ITS '92 Montreal, Canada, June 1992 proceedings. Springer-Verlag. Berlin Heidelburg.

Asymetrix Corporation (1989) Using Toolbookreg.: A Guide to building and working with books (Version 1.5). Washington

Brown, J. S. (1990). Toward a new epistemology for learning. In Claude Frasson and Gilles Gauthier (Eds.) Intelligent Tutoring Systems. New Jersey: Alex Publishing Corp.

Carlson P & Gonzales G (1993) The knowledge board: Using hypertext as an "intelligent" workspace for writing issues-based prose. Journal of educational multimedia and hypermedia. Vol 2 No.14, p 417-431.

Chan T, Lin C, Lin S and Kuo H (1993) OCTR: A model of learning stages. In Brna P, Ohlsson S and Pain H (eds) Proceedings of AI-ED 93 . Scotland, August 1993. ACCE publication.

Cobb, P., Wood, T., Yackel, E. & McNeal, B. (1992). Characteristics of classroom mathematics traditions: An interactional analysis. American Educational Research Journal, 29, 3, 573-604.

Corbett A, Anderson J, Patterson & Eric G (1990) Student modeling and tutoring flexibility in the lisp intelligent tutoring system. In Claude Frasson and Gilles Gauthier (Eds.) Intelligent Tutoring Systems. New Jersey: Alex Publishing Corp.

de La Passardiere B & Dufresne A (1992) Adaptive navigational tools for educational hypermedia. In I Tomek (ed) Computer assisted learning. Proceedings of the 4th International conference, ICCAL '92 Wolfville, Novia Scotia, Canada, June 1992. Springer-Verlag Berlin Heidelburg.

Duchastel P (1992) Towards Methodologies for Building Knowledge-Based Instructional Systems. Instructional Science, v20 n5-6 p349-58.

Eklund J (1993) Cognitive modeling in intelligent tutoring: Individualising tutorial dialogue. In Australian Educational Computing. Proceedings of ACCE Conference, Sydney 1993.

Jonassen D (1992) Designing Hypertext for Learning. In Scanlon E & O'Shea T (eds) New directions in educational technology. Springer-Verlag. Berlin.

Jonassen D (1991). Objectivism versus constructivism: Do we need a new philosophical paradigm? Educational Technology, Research and Development, 39, 3, 5-14.

Karlgren J, Hook K, Lantz A, Palme J & Pargman D (1994) The glass box user model for filtering. Poster session presented at the Fourth International Conference on User Modeling, Hyannis, Massachusetts, USA, August 1994. Mitre Corporation.

Kobsa A, Müller D & Nill A (1994) KN-AHS:AN adaptive hypertext client of the user modeling system BGP-MS. Fourth International Conference on User Modeling, Hyannis, Massachusetts, USA, August 1994. Mitre Corporation.

Knussen C et al (1991) An approach to the evaluation of hypermedia. Computers and education, Vol17 No1 p13-24.

Messing J (1990) The use of content and teaching strategy control features in computer assisted learning courseware. Charles Sturt University. Riverina. (ERIC document ED 59246)

Misanchuck E & Schwier R (1992) Representing interactive multimedia and hypermedia audit trails. Journal of educational multimedia and hypermedia. Vol1 No3 p355-72.

Nelson T (1978) Dream machines. South Bend, IN: The distributors. Cited in Jonassen D (1992) Designing Hypertext for Learning. In Scanlon E & O'Shea T (eds) New directions in educational technology. Springer-Verlag. Berlin.

Ohlsson S (1992) The interaction between knowledge and practice in the aquisition of cognitive skills. Pittsburg University, Pa. Learning and development centre.

Pitz H (1994) "Adaptive customisation" of user interfaces. Fourth International Conference on User Modeling, Hyannis, Massachusetts, USA, August 1994. Mitre Corporation.

Santiago R & Okey (1992) The Effects of Advisement and Locus of Control on Achievement in Learner-Controlled Instruction. Journal of Computer-Based Instruction, Vol19 No2 p47-53.

Van Lehn K (1992) A workbench for discovering task specific theories of learning. In New directions in educational technology. Scanlon E & O'Shea t (eds). Springer-Verlag, Berlin Heidelberg.

Vassileva J (1994) A Practical architecture for user-modeling in a hypermedia-based information system. Fourth International Conference on User Modeling, Hyannis, Massachusetts, USA, August 1994. Mitre Corporation.

Yazici H, Muthuswamy K & Villa J (1994) An intelligent system approach for graphical user interface management. Journal of educational multimedia and hypermedia. Vol 3 No.1, p 37-54.


© Southern Cross University, 1995. Permission is hereby granted to use this document for personal use and in courses of instruction at educational institutions provided that the article is used in full and this copyright statement is reproduced. Permission is also given to mirror this document on WorldWideWeb servers. Any other usage is expressly prohibited without the express permission of Southern Cross University.
Return to the AusWeb95 Table of Contents

AusWeb95 The First Australian WorldWideWeb Conference