Adaptative Hypermedia: Supporting the Communication Process

Aude Dufresne

Departement des Sciences de la Communication

Universite de Montreal

Case postale 6128, succ Centre-Ville Montreal, Canada H3C 3J7

Tel : 514 343 7371, Fax : 514 343 2298

e-mail :

Why should the interface be adaptative?

Two applications were described for hypermedia (Berk, & Devlin, 1991). First, they may serve as a thinking tool, to organize and facilitate access to personal and known information; second it can be a medium by which the user access information organized by others. The necessity to develop adaptative interface is relevant mostly in the second type, because then the user is confronted to a vast set of documents of which the structure and content is unknown.

In this context, the function of the hypermedia is to communicate, making the information accessible and useful to the user. To understand the difference between presenting an information and making it accessible, it is interesting to think of the interface as a dialog, where the information is exchanged in both directions, where meta-communication and non-verbal cues serve as non-intrusive feedback to guide the exchange, finally where the information is gradually adapted to the level of competence and interest of the interlocutor.

Paradoxically, adding intelligence and control in the interface is contrary to the philosophy of hypermedia which are supposed to give the user full control to explore the content at a given moment. More so, an adaptative interface may be perceived by the user as unpredictable and incoherent, increasing his impression of disorientation. Therefore, it is important to design the adaptation to make it both predictable and flexible to the user.

We will discuss the problematic of adding adaptative interfaces to hypermedia and present how historical information may be used to modify the interface in a non-intrusive and non-disorienting way. We will first present types of information on the navigation, that can be used for this adaptation and then we will show how the interface can be adjusted to help the user.

Building a user model to adapt the interface

First, a model of the user and his progression must be constructed so that the system can adapt using it: displayed information, competence demonstrated in tests, types of errors can be used to determine potential knowledge of the user. For these both local and global measures may prove to be pertinent, for example three errors consecutive on the same test vs three consecutive errors on different tests vs three non-consecutive errors may trigger different types of feedback or adaptation of the system. But also, the patterns of progression as manifested in the frequency of consultation, in the time of consultation, in the schema of navigation (depth first, breath first, browse-then read, etc.) can also be used to adapt the interface.

Adapting the interface

The adaptation in the interface must appear as a continuous process; it must be translated at the meta-communication level, so the user notices what is happening. He must be able to distinguish the static structure of the content from the dynamic properties associated with the adaptation.

The idea to have intelligent agent in an interface is becoming very popular; programs, like Sesame or Excel 5.0 are programmed to notice recurring actions of the user and to propose taking charge of it: "Do you want to throw away all the files in the current folder ?" Direct offering to the user is one way to both acknowledge and verify with the user the pertinence of an intelligent adaptation. But this kind of intervention interrupt the current cognitive process, it is situated at a level of abstraction and of distanciation from the actions, the user is not always ready to understand. So we will try to define other ways to support the user in a more natural and less intrusive way.

A more passive type of adaptation is to give the user an intelligent feedback on his activity. As presented in the context of education (De La Passardihre, & Dufresne, 1992; Dufresne, 1992; Dufresne, Jolin, & Senteni, 1990), contextual and historical cues in an interface are very effective in guiding the user in his exploration.

Contextual cues

Contextual cues relieve the user cognitive load in short term memory, by presenting him in a concise way the context in which he is looking at an information: presenting and giving direct access to recent information (title and subtitle, title of recent pages, near subjects of interest, relative position toward completion) are ways to help the user remember what he is doing. These may be highly contextual, so that they vary depending on the path of access to a specific information. In the same way, Mayes, Kibby, & Anderson (1990) describe the use of "signposts" to support orientation in hypertexts. Maps or graphical browser also support orientation and suggest path of exploration. Contextual cues introduce redundancies which facilitate understanding and remembering of the content.

In the context of adaptation it is possible to augment the accessible information gradually, adding complexity in the graphical browser, as in zoom in or fisheyes views; the content being gradually expanded into new areas to be explored. The concept of a genetic organization of knowledge described by Goldstein (1982) can be implemented in an hypermedia, making the exploration of information similar to an adventure game, where the user have to gain access to levels of information. This augmentation have to appear continuous as in a spiral so that the user is expecting and looking forward to it. The general structure have to stay the same, while elements are simply added to it.

Historical cues

Historical cues are also very useful, they translate to the user his progression in the content. Various historical cues have been proposed, which are marks in pages or in a graphical browser to indicate that a user have seen an information. We have explored historical cues (Dufresne, 1992) and we have found that it is important to offer progression cues in menus, when the user is choosing a path ( vs in a general browser or after he has chosen). Also those progression cues have to be nuanced, so the user can differentiate how deep in a subject he have explored.

In this research, we have found that offering a three stages progression cues (not explored, partly or completely explored) was sufficient to increase significantly the exploration of a hierarchical content. The interface can thus be used in a non-intrusive way, to give the user information on his progression at the meta-communication level.

Knowledge of the user's progression can also be used to adapt reactions of the system to the user's errors or hesitations. Information on the number of times a user have tried to do an operation should be used to modify the feedback.While the user, trying to understand, repeats a manipulation (looking for an information in the wrong place, giving the wrong answer); he will react negatively, if the system does not acknowledge this repetition by answering differently. To follow the dialog analogy, the system have to change his error messages depending on the number of errors repetitions. For example, he may suggest similar information or a different access mode, if the user keeps entering and exiting the same menu choice.

We have found that historical cues have to be cognitively coherent, reflecting the extent or exploration of a subject, independently of the mode of access to the information, i.e. an information have been seen, whether through the index or regular browsing. To establish this cognitive coherence, the adaptative interface have to be based on a generic model of the information; the user model is then an annotation of the user's progression in this structure, We have shown how a distributed model of the content and of the user progression can easily be implemented in an Hypercard application (Dufresne, & Tremblay, 1991). Which aspects of the historical progression are to be translated to the user may depend on the domain, for example in the area of learning, historical cues may reflect predictions of the understanding of the user: "You must know this, since you have demonstrated this knowledge", "This should be harder for you to understand".

Finally, historical information may also be used to modify the level of user control on the system. Even in an exploratory system, it is sometime interesting for the system to take control of the interaction. Turn taking may be imposed with or without permission, if the user appears to be in a dead end (ex: a sequence or loop is repeated more than twice, a time limit is exceeded).Turn taking may also be offered as an expert or guide, that the user chooses to consult. The system may then give the right answer or present pertinent information; he may restrict or augment the user mobility in the system ( 3Do you want to start again?2or 3Do this point first2 may be suggested, if the user have made many errors). If is behavior appears disorganized, a walk through or demo may be suggested.

A very promising area of research, is to use the historical model of the user's progression to recognize cognitive styles of exploration and to support them. For example, speed of browsing, number of menu items, number of examples can be adjusted depending on the user's previous manipulations.


Berk, E., & Devlin, J. (1991). Hypertext/ Hypermedia Handbook. New York: McGraw-Hill, 583.

De La Passardiere, B., & Dufresne, A. (1992). Adaptative Navigational Tools for Educational Hypermedia. In I. Tomek (Ed.), Computer Assisted Learning (pp. 555-567). Berlin, New York: Springer-Verlag.

Dufresne, A. (1992). Ergonomie cognitive, Hypermedias et Apprentissage. In B. d. LaPassardiere, & G. L. Baron (Ed.), Actes des premieres journees scientifiques Hypermedias et Apprentissage (pp. 121-132.). Chatenay-Malabry, Paris:

Dufresne, A., & Tremblay, I. (1991). Modelling Distributed Knowledge to Support Learning Environments in Physiology and Computer Science. International Conference on Simulation in Engineering Education. Simulation Series, 24(2), 185-189.

Dufresne, A., Jolin, N., & Senteni, A. (1990). Hypertext documents for the learning of procedures. In R. M. ALEESE, & C.GREEN (Ed.), Hypertext: State of the Art (pp. 96-104). New York: Ablex.

Goldstein, I. P. (1982). The genetic graph: a representation for the evolution of procedural knowledge. In D. Sleeman, & J. S. Brown (Ed.), Intelligent Tutoring Systems (pp. 51-78). New York.: Academic Press.

Mayes, T., Kibby, M. R., & Anderson, T. (1990). Signposts for conceptual orientation: some requirements for learning from hypertext. In R. M. ALEESE, & C.GREEN (Ed.), Hypertext documents for the learning of procedures