Contributions of This Thesis

Contributions of This Thesis

To avoid some of the problems outlined above our approach to designing, implementing and evaluating the POP system was to start off from users’ real needs and behaviour. We utilised a human-computer interaction approach, which we take to mean starting from users’ needs and repeatedly evaluating the developed tool with real users. In this approach, we have utilised task analysis as a method for getting closer to what users’ real problems and needs are, and rapid prototyping as a means to evaluate our ideas for design. In particular, we have been concerned with the problem of bootstrapping the adaptive behaviour, which we have tackled through several small evaluations of the rules for adaptation.

We also had a strong focus on finding an adaptive behaviour that would substantially improve users’ interaction with the system. In our domain, the two most prominent problems were information overflow and, once the information had been found, interpretation of the information found. Instead of making small, subtle changes to the information by, for example, avoiding concepts unknown to the user, we aimed at making substantial, but conceptually simple adaptations where whole pieces of texts are hidden from the user’s immediate view. Rather than adapting to users’ knowledge, we try to adapt to their information-seeking tasks. If we know that the user is trying to find information that would help him/her in, for example, a project planning task, we hide all the information which is irrelevant to this task. Thereby, we attempt to get at the underlying needs of users rather than only focus on their knowledge, as well as making quite robust and useful adaptations. Furthermore, we shall show that adaptation to users’ tasks makes it much easier for the author of a text to shape its content to the future reader – a crucial requirement if our adaptive system is going to be easily maintained by the authors of the information.

By focusing on users’ information-seeking tasks, we also focus on short-term characteristics of the user as a basis for adaptation. We do not have to make guesses about users’ long-term characteristics (like their knowledge), but instead we adapt interactively during one session with the system and faulty assumptions made will have less impact on the system’s overall reliability. Our adaptations are also such that we do not take away information completely, we only hide it and the user can always choose to open the hidden information.

Our work has been much concerned with ways in which the user can control the adaptivity. Our design has been inspired by the ”black box in a glass box” metaphor put forth by du Boulay, O’Shea and Monk, (1980), in the area of design of programming language environments. In order to make it possible for the user to control the adaptivity, we present a simplistic, domain-dependent, view of what the adaptivity does. We also try to display the results of the adaptive system in a visual manner. The user is then allowed to change any assumptions made by the system. The complex behaviour of what is really going on in the adaptive system is hidden from the user in the ”black box”. This can be compared with what most programming language environments do when they allow the user to trace a program: the trace reflects some abstract machine that simulates the behaviour of the program – what is really going on, in all its gory details, is not always shown to the user, (Höök et al. 1990).

The techniques we utilise for realising our adaptive system fall under the heading adaptive hypermedia (Böcker et al. 1990). Adaptive hypermedia will follow the user’s actions at the hypermedia system and actively try to adapt either (1) the content of a page of information, or (2) what clicking on a hotword³ should result in, i.e. the navigation in the information space. We realise our hiding of information through a stretchtext technique: by clicking on a header of a piece of closed text, the user causes the text under the header to be inserted into the description. When our system adapts it will choose which headers should be opened and which should be closed. As a minor adaptation, the system will also choose to either use abbreviations or the full name of concepts in the target domain. We attempted to adapt the navigation, but only in minor ways. Our approach to adaptation will therefore mainly fall under (1): adaptation of the information content on the page.

In order to do the kind of adaptation that we did in our system, the domain must be such that it is possible to organise the concept domain in sets of similar ”concepts” which are presented by one description or ”page” each. In that page there must be several headers under which different aspects of the concept are presented. If this is possible, many advantages can be gained:

• 
  We can set up relevance criteria for each set of concepts which determine what information should be shown when, i.e. which headers should be opened and which should be closed based on the assumed intention (or task) of the user.
  

• 
  We can know beforehand how to interpret the actions of the user: if a certain header is opened by the user, we know what kind of information is requested and we can start attributing intentions to the user.
  

• 
  We can decide on an organisation of the database that is optimal and then create powerful tools by which the authors of the information can input the information about the concepts.
  

On the interface level, there are some interesting aspects of our system:

• the interface is multimodal: both text and graphics are generated or retrieved in response to the user’s queries, and both clicking and queries are allowed as input from the user.

• the interface is implemented in WWW using html and Java in a novel way, with stretchtext, hotlists⁴, menus and local maps of the information space.

• the adaptivity is visually displayed in a way that enhance users’ learning of the relation between the system’s assumption about users and the corresponding adaptation – thereby improving the possibility to learn about and control the adaptive behaviour.

As a side effect of our studies of the user’s problems with the information overflow and navigation in the hyperspace, we also studied other aspects of human cognition that either should be catered for by an adaptive behaviour of the system or by careful adaptable⁵ design. We found that an individual user’s spatial ability determined how well s/he could navigate to and retrieve information in the large information space of this domain. But instead of trying to infer users’ spatial ability from their interactions with the system and adapt to it, we decided to design the interface to allow for several different ways of navigating and posing queries. Users can choose their preferred interaction style, and, to some extent, change the interface.

Finally, an important contribution of this thesis is the comparative study we made of two variants of our system: one adaptive and one non-adaptive variant. As said above, there are not that many studies of adaptive systems, and, of course, even fewer studies of adaptive hypermedia systems. In the study, the subjects clearly preferred the adaptive system. Also in more objective measures, we could see that subjects’ information-seeking behaviour was improved as the adaptive system required fewer actions within the page, and the choice of information made by the adaptive system influenced subjects’ solutions.