Despite their increasing appeal, adaptive systems in general and adaptive VEs in par- ticular pose several challenging problems to their designers. Tracking user behaviours, analysing and extracting relevant behavioural patterns, identifying the interrelation- ships between these patterns which should become part of a model aimed to explain and predict user behaviour represent aspects which require a great deal of work. Once the user model has been identified, one should search for possible ways to address the most important aspects embedded in this model which can potentially increase task performance. The real time constraints for providing adaptivity, and the additional difficulties of predicting the dynamics of user behaviour are aspects which depict the problem domain of building such adaptive systems.
2.6.1 User Adaptive Systems
Jameson (2003, p. 317) defined user-adaptive system as follows:
An interactive system that adapts its behaviour to individual users on the basis of processes of user model acquisition and application that involve some form of learning, inference or decision making.
The general schema of a user-adaptive system (Jameson, 2003) is presented in Fig- ure 2.2, where ovals represent input or output, rectangles represent processing methods and the grey 3D figure represents stored information.
Langley (1997) defined adaptive user interface as “an interactive software system that improves its ability to interact with a user based on partial experience with that user”. Emphasising the concept ofadaptability as a feature of systems able to cope with dynamically changing user requirements, Savidis et al. (1997) noted that it reflects a
User Model User Model Acquisition User Model Application Information about User Predictions about User
Figure 2.2: General Schema for User Adaptive Systems
“capability to automatically tailor itself initially to each individual end-user”. These definitions emphasise two key aspects which distinguish adaptive systems from other intelligent interfaces:
1. The central role of user modelling. This involves acquiring information about the end-user, usually by employing machine learning techniques (see Chapter 6). The main outcome of this stage consists of identifying the users’ patterns of behaviour and drawing assumptions about them. As Kobsa (1994) noted, the systems should also be capable of storing these assumptions appropriately and inferring newones. 2. Tailoring system behaviour to the end-user. This involves comparing current user’s behaviour with the model previously acquired and deciding how to adapt the system in order to meet user’s perceived needs.
Jameson (2003) identified two major purposes of user adaptive systems in terms of supporting system usability and user modelling. He detailed these two goals by nine functions:
• Supporting system use:
– taking over parts of routine tasks,
– adapting the interface,
– giving advice about system use,
– controlling a dialog;
• Supporting information acquisition:
– helping users to find information,
– tailoring information presentation,
– recommending products,
– supporting collaboration,
Adaptive Virtual Environments
VEs feature an appealing potential not only for studying spatial skills and abilities but, more importantly, for training them. The well-known large inter-subject variability characterising these skills raises the issue of their proper assessment. In this respect, VEs have a lot to offer, given their capacity of providing frames for the applied spatial tasks as complementary assessment tools to traditional tests of spatial skills and abilities. An interesting aspect in this context relates to the limited predictive validity of these tests with respect to performances on spatial tasks in VEs. One major reason which explains this refers to the fact that whereas tests seem to focus on spatial properties of objects, the tasks focus holistically on spatial properties of environments (Durlach et al., 2000).
As Durlach et al. (2000) noticed:
Not only can this technology (VE) be used to provide a wide range of basic spatial skills and abilities training exercises . . . , but it is also capable of providing a complete and real time continuous record of trainee responses and thus provide immediate feedback.
Given both importance and prevalence of navigational tasks within VEs, surpris- ingly little effort has been invested in the development of adaptive VEs in order to accommodate individual differences which lead to significant difference in user perfor- mance. Navigation assistance in VEs is a necessity, given the difficulties encountered by lowspatial users during their navigation (Benyon, 1993).
To the best of my knowledge, there is only one example of an adaptive virtual environment developed for supporting navigation, at whose core lies a user model of navigation. However, this system is not a VE such as that described in Section 2.3, based on a physical metaphor, but a hypermedia VE or so-called information system. This model led to the identification of two strategies employed during search in abstract spaces, strategies mapping individual differences in the style of processing information (Ford et al., 1995; Ford, 2000; Chen and Ford, 1997). The holistic strategy enables a global perspective of the content being learned, aiming to build a broad conceptual overview into which details will fit later, whileserialist strategy enables a local learning, based on a thorough examination of individual topics with the overall picture emerging later. Such an intelligent adaptive system can identify user’s learning strategies, classify these strategies in terms of a learning model, offer individualised learning support, and use the feedback for improving the model over time. A reviewof other attempts of building adaptive VEs for training navigation is presented in Section 3.3.3.
Given the central role played by user model in user adaptive interface, the following subsection introduces some relevant issues in the area of user modelling. Referring to navigation in information systems, Ford (2000, p. 543) noted:
Current technology allows the development of information systems that offer flexibility in terms of routes through subject content and a rich set of navi- gational tools enabling varying levels of user and program control. However,
we urgently need robust user models to enable us to optimize the deployment of such facilities. Research into individual differences suggests that system efficiency and effectiveness may be enhanced by adapting to individually different needs on the part of users.
2.6.2 User Modelling
User modelling is a growing discipline in the field of HCI, extending itself in various areas which focus on the development of user adaptive systems. The major reason for this resides in the fact that these systems are and will continue to be used by heterogeneous user populations. In order to adapt themselves to the end-user, systems must be able to make assumptions about their users, relevant for tailoring their behaviour to the users (Kobsa, 1994; Fischer, 2001).
The field of user modelling benefits from methods developed in other areas, for example machine learning, knowledge representation, and HCI, adapted to the specific needs of this field (Kobsa, 1994). For a detailed presentation of some of these techniques, see Chapter 6.
“A distinctive feature of an adaptive system is an explicit user model that represents user knowledge, goals, interests and other features that enable the system to distinguish among different users” (Brusilovsky and Maybury, 2002).
A user model is a model that a system has of its users. Such a model resides inside a computational environment, as opposed to the mental model, developed by users with respect to the systems and tasks, and which resides in users’ heads (Fischer, 2001; Norman, 1983; Finin, 1989) (see Section 3.2.1). The impact of user mental model on successfully performing tasks with any kind of system raises the need of taking this model into consideration for the purpose of system design. This would ensure increased system usability (Preece, 1993). According to their purpose, user models could be used to describe (Webb et al., 2001)
• the cognitive processes that underlie the user’s actions;
• the differences between the user’s skills and expert skills;
• the user’s behavioural patterns or preferences;
• the user characteristics.
Users often encounter difficulties when navigating in unfamiliar physical places, and in particular in unfamiliar VEs (Vinson, 1999; Waller, 2000; Dijk et al., 2003). The computational complexity characterising VEs imposes limitations on the level of visual details accommodated by virtual worlds (Ruddle et al., 1997). This is expressed in terms of fewer landmarks and depth cues (Vinson, 1999). The problems encountered in VEs are primarily related to limited sensorial stimulations. This applies especially to desktop VEs because of their drawbacks such as restricted view field (Waller et al., 1998),
absence of peripheral vision (Ruddle et al., 1997), the difficulties of depth perception, restricted kinaesthetic (Waller et al., 1998) and proprioceptive inputs (Ruddle et al., 1997), and motion sickness as a negative side-effect of navigating in VEs (Harm, 2002). These difficulties of navigating in unfamiliar VEs suggest the need to support navigation in VEs (Vinson, 1999). One way to address this is through accommodating users’ individual differences.
2.6.3 Accommodating Individual Differences
Accommodating individual differences in navigation related areas can benefit from the methodology proposed by Egan and Gomez (1985), based on Messick’s work (1976) on accommodating individual characteristics for instructional process (Chen et al., 2000). This involves three strategies such as the challenge match, the capitalisation match and the compensatory match.
• The challenge match is generated by high task demands which strain users’ ca- pabilities and force them to adapt. It works well when user abilities are properly estimated and the challenge is not too high to lead to failure or lack of motivation.
• The capitalisation match involves harnessing the user potential, in terms of knowl- edge and skills, through properly tailored tasks which will not exceed these capa- bilities.
• The compensatory matchaddresses particularly user’s weaknesses and limitations through additional help such as training, assistance or mediators unavailable by default.
For instance, Stanney and Salvendy (1995) stressed the significance of the spatial mental model of navigating in abstract informational spaces, as a key task component that caused the differences between high and low spatial individuals’ task performance. By eliminating the need to mentally visualise the structure of the information, low spatial individuals were able to perform as well as high spatial individuals. In this case, the interfaces successfully compensated lowspatial users.
The compensatory match has been further developed, throughsupplantation which entails engaging in the activity on behalf of the user so that the required level of compe- tence is reduced, and throughfacilitation which consists of providing tools for enhancing user’s engagement in solving the task, i.e. feedback, appraise, validation, review(Ford, 2000).
Investigating possible ways of addressing cognitive style-related individual differ- ences in searching hypermedia, Ford (2000) suggested three access modes to enable differential navigational patterns:
• Autonomous access, which enables users to choose access patterns for themselves.
• Prescribed access, which requires both a mechanism for assessing the end-user competence and a model robust enough for generating an effective access pattern for each user on the basis of such an assessment.
• Recommended access, which offers a default navigational access mode, possibly overridden by the user.
Egan and Gomez (1985) suggested a three stage approach to accommodate individ- ual differences: isolation, assaying, and accommodation. These stages relate to user, to task and to the match between user and task (Chen et al., 2000)
• Isolation consists of identifying those individual differences with maximal impact on task performance.
• Assaying involves the identification of those components of task that account for performance variability.
• Accommodation consists of altering previously identified key task components in order to adapt the task to user capabilities according to one of the three matching strategies.