The term ``user model'' as well as ``user modeling'' is overloaded with different meanings in research literature. As a general definition, a user model is a computer system's model of user characteristics for the purpose of tailoring the interaction or making the dialog between the user and system adaptive [Murray, 1987]. However, user characteristics can have many dimensions: knowledge about the computer system, knowledge about the domain, goal of the current task, preferences, cognitive and learning abilities or disabilities, and so on. When the user model represents the goal of the current task, it overlaps with the task modeling described in Section 5.2.1. Furthermore, a user model has different temporal dimensions [Dieterich et al., 1993]. When it is used to describe characteristics of a user valid only in the current context or session (short-term data), it overlaps with the discourse model described in Section 5.2.3.
Throughout this thesis, the term ``user model'' is used exclusively in the following sense. A user model represents the background knowledge that a user possesses and is kept as long-term data on a permanent storage medium. As shown in Figure 4.1, a user's knowledge about an information repository falls into four levels. A user model should contain pieces of information falling in both L1 and L2. Because information belonging to L1 is well known and regularly used, there is no need for the system to actively deliver it. Although information belonging to L2 has not been completely acquired by the user yet, it can still be considered as a part of the user's active knowledge because the user knows about it and will use it readily when it is needed. Even if the user may need more details about it, he or she knows very well how to find them with information access mechanisms. Accordingly, user modeling is used in this thesis to refer to the computational mechanism embedded in the information systems for the creation, augmentation, and maintenance of such user models.
Similar to discourse models, user models can be explicitly modified by users or implicitly updated by the system. Direct modification from users requires that the system be adaptable, which means users can customize the system behaviors to their own needs. Adaptive systems automatically update user models based on information observed or inferred from monitoring their interactions with the system [Fischer, 1993].
Adaptability and adaptivity complement each other. Although adaptivity requires little effort from users, it needs a relatively long time to establish a reliable user model. Deployment of VDDE, an active design environment for phone-based interface design, has found that experienced designers do not expect to be interrupted with information they have told the system irrelevant [Sumner, 1995]. Adaptability of systems gives users direct and immediate control over what information should be delivered. However, it places extra work on users.
Another challenge in the acquisition of user models is how to initialize a user model. Few users know nothing about an information system when they start to use it, and, obviously, empty user models do not reflect this fact. Mechanisms supporting adaptability and adaptivity can be extended to the explicit and implicit acquisition of initial user models, respectively. An explicit acquisition method directly asks users what they already know through an up-front questionnaire or testing sessions when they use the system for the first time. An implicit acquisition method is suitable if artifacts previously created by users in the domain for which the information system is designed to support are available. The interactive adaptivity mechanism can be modified as a batch process to analyze those existing artifacts to obtain the initial user models. The third method is to create several stereotypical user models to represent different levels of users, as is widely done in intelligent tutoring systems and intelligent help systems. One of the stereotypical models can be chosen as the approximation of the initial user model, or, a user can copy the user model of another user who has a similar level of knowledge.