Abstract: In this paper, we present some developments of CHEOPS , a package that can be used by WWW designers to provide their hyperdocument with adaptivity and navigational aids.
Previous version of CHEOPS , presented in (Ferrandino et al. 1997), is implemented by several CGI-BIN scripts that enables an HTTP server to interact with a user and adapt its responses to her behaviour. Solution is transparent to the user and to the client, and has a moderate impact on server performances.
What we present here is the extension of CHEOPS that takes into account user modeling (i.e. extends adaptivity to multi-session interactions) providing also annotations capability. The new version of CHEOPS , named 1.0, is developed in PERL and is therefore more portable.
We conclude by sketching future directions of CHEOPS and some possible usage of the knowledge model in on-line help and training systems.
Keywords: User Adaptivity on WWW, PERL, CGI-BIN
Among them, one of the most recognizable is the educational usage of WWW. In fact, several educational systems based on WWW have been developed in the recent past and many results are available on the educational use of the Web for distance learning [Dwyer et al. (1995),Hammond and Allison (1989),Ibrahim (1994),Ibrahim and Franklin (1995)]. In particular, in [Ibrahim (1994)] the need for a ``WWW work session'' (using our terminology) in an educational setting was recognized and an application based on CGI-BIN programs was developed that adds a similar capability to the Web server.
In order to fully exploit WWW potentiality in the educational field, server response must be adaptive: it must, in fact be aware of user's behaviour so that it can take into account the level of knowledge and, as a consequence, provide the user with documents that she can read, given her background/capacity.
Adaptive hypermedia is a recent research area whose goal is to enhance the functionality of hypermedia by building a model of the user and adapting the response accordingly[Brusilovsky (1996),Brusilovsky (1997)].
What we propose here is a system that is based on an ``opaque'' mechanism, that, based on the history of user's requests, can provide a different document that is likely to be understood and lightly challenging while not being intimidating. This can be useful also in other contexts, i.e., avoiding that the large amount of information available can scare a user off the site, since she can `` have trouble in finding the information they need'' [Nielsen (1990)] and, therefore, can loose interest in the site itself.
CHEOPS is a design system that should make easy for a designer to add adaptivity to a hyperdocument in a modular way. Especially with the extensions described here and available in version 1.0, this goal looks quite reachable.
CHEOPS adopts the following strategy: each time a new user is presented to the systems, the server creates a user's profile, ``distributing'' (virtually) a ``session card'' embedded within the hypertext file that is sent back to the browser. Each successive requests to the server will be done with this session card and, therefore, recognized as belonging to a specific session. The system is also able to recognize (by a username-password pairs) users and recover the profile and the history for her.
CHEOPS system is non-obtrusive, widely usable, (one does not need specific browsers to use our system), has moderate impact on server load (putting all the ``memory'' of the session where it belongs, i.e. on the disk) and is modular enough to be easily installed on a Web server for a specific hyperdocument.
CHEOPS is implemented through several CGI-BIN PERL scripts, used to build ``on-the-fly'' documents that depend on user's previous inquiries at the same server. CGI-BIN scripts are program run on the server and triggered by input from a browser.
The system, given a hyperdocument as HTML pages, is able to provide the requested page to the user according to her profile that is modified each time the user chooses a link. The system provides the page adding additional information like links to context-sensitive help, information about her current profile (given in terms of knowledge levels for each category, details follow in the next subsection), links to ``summaries'' for each category, a form to add annotation to each visited page, link to the history of the visited documents and so on.
CHEOPS version 0.9 (presented in [Ferrandino et al. (1997)]) consisted of several Unix Shell scripts that provide the designer of a hyperdocument with an automatic mechanism that satisfies user's requests according to previous inquiries. The extension presented in this paper are mainly about the user profiling mechanism and the annotation capability that has been added to the system.
Analysis of requirements. Being focussed on developing a design tool for hyperdocument to be employed in the educational field, several requirements were identified during preliminary experiences:
- Lecture-type Interaction: The user should be able to interact
with a WWW hyperdocument that makes her feel like having a
teacher/instructor/field-expert `` on the other side of the
screen''. To reach such a goal, it is necessary to have tools
able to distinguish between requests of different users. It is also
recommended the usage of context-sensitive help, where by context it
is not only meant the part of hyperdocument currently visited but
also the user knowledge and previous interactions with the help
- Categories: It is often the case that there is a ``natural'' subdivision of the hyperdocument in categories that are well-suited to the argument. It is important that user is able to check her ``knowledge level'' for each category and also able to choose to further explore a category that, willingly or not, was ignored at a previous step.
- History Mechanism and Annotations: User should be able to check the path followed during her interaction with the hyperdocument: being able to trace back previous interactions with the comments to each single visited page is extremely helpful in developing and reviewing the actions taken.
The Session Protocol. The session protocol that has been implemented is based on the following sequence of actions:
When a hyperdocument is designed for a broad audience, it is customary to face several problems in trying to make its style acceptable for novices, amateurs and experts at the same time. They all have different, some time opposite needs: a novice may need information that can be boring for an expert since it makes her loose time in finding the more advanced part of the hyperdocument. On the other side, a novice can be scared and confused by an excessive usage of technicality and leave the site before she could have the chance to access information that are suitable for her experience in the field. Then, great care must be taken to fulfill this important, we dare say paramount, requirement if one is asked to use at its best such an efficient communication media like the World Wide Web.
Another orthogonal need is to help the user (whichever level and experience they have) in navigating through the (sometimes huge) amount of information. This is often achieved by subdividing data in a limited number of categories that can be seen as providing an hypertext of a small degree of traditional linearity within the hypertext itself.
Putting all together, the view we propose as a model for dealing with the ``shapeless'' amount of information is a geometric one: let us build a Knowledge Pyramid, having as bases two regular polygons of the same shape but of different size, the smallest on the top. Each vertex of the polygon represents one of the categories which the designers subdivided the whole amount in; the edges of the pyramid represents the experience level in each category.
The experience acquired by a user is defined as the surface obtained interpolating the intersection of the current experience levels on each category. The learning process can be seen as trying to get the surface as low as possible.
A possible variation of the model can be to have fewer category on the top than on the bottom. This corresponds to the natural mechanism to avoid mentioning some categories if the user may not be ready and actually may be disturbed by that.
CHEOPS uses the Knowledge Pyramid as an important navigational aid to the user: at each step, the user is able to see what her confidence for each category is (or is supposed to be by CHEOPS ) and act as a consequence. She may choose to explore a part of the hyperdocument she left behind, or she may want to go deeper in a category or she may go on with her hypertextual visit of the document.
The entire package has been moved from csh to a more portable and powerful language as PERL. The system, moreover, uses Relational Database System ([RDB]) for storing both users' profiles and history information.
Annotations. CHEOPS 0.9 provided a history of the interactions to the user. Now, the system allows the user to add annotations to each visited page, by evaluating each page as ``Not interesting'', ``Interesting'' and ``Very Interesting'' and by adding a one-line comment. At this moment, annotations are visible only to the user that wrote them. Future developments will allow to share comments on pages.
Adaptive and Adaptable. The system now fully allows the user to choose and modify the knowledge levels. CHEOPS is, therefore, not only adaptive but also adaptable, a characteristic that is very important for experienced users.
Categories. In this version, we have enhanced the way the designer can choose to adapt CHEOPS to its domain-specific application. In fact, while inCHEOPS 0.9 only a fixed number of categories were available, in the new version the designer can choose the number of categories as she likes. Moreover, and maybe even more important, the designer can decide to have a multi-faceted pyramid: a category can appear to the user only if a certain knowledge level on other categories has been reached.
User modeling. The system is now able to keep the current profile of each user in a DataBase so that it can be easily retrieved in successive interactions. The system allows also (on a WWW server Apache [Apache]) to insert and change password for users through a fill-in form.
Removable Headers/Footers. CHEOPS inserts header and footer for each page. Now the designer can choose to provide or not headers/footers for each page while the user at run-time can choose to turn headers/footers on and off.
We feel that the knowledge model introduce by CHEOPS can be very helpful in fields that are only marginally related to ``pure'' educational purposes as in Computer Based Training (CBT) systems or on-line help and training systems.
In this case, in fact, the CHEOPS knowledge model can be extended to encompass two pyramids, a top one and a bottom one, connected as shown at the left. The motivation is that, at beginning, the trainee is in ``training'' process and, therefore, needs a lot of information. When the trainee is sufficiently expert she does not need such a huge amount of information and therefore, her profile is now staying in the area between the bottom and the top pyramid. If necessary, when some particular problem requires extremely specific information, the trainee can go deeper and deeper and her profile moves down to the bottom pyramid.
Future extensions of CHEOPS are, therefore related to such applications. We plan to study the training process and develop different knowledge models by providing, at the same time, a package that allows to realize such adaptive systems on the World Wide Web. Relevant future work include dealing with multiple user annotations and providing adaptivity in a cooperative environment, where different trainees collaborate over the learning process and are seen by CHEOPS as a group of users that, as a whole, behave as a unique user still mantaining personal peculiarities. Moreover, research on authoring systems for CHEOPS is also planned.