Selection of the appropriate teaching strategy for a particular user is a difficult task. Several ITSs use complex student modelling and diagnosis techniques to try to provide an accurate view of the user and to allow teaching strategies to be modified accordingly. Examples of this can be seen in the previously described systems DOMINIE (Spensley, et al., 1990) and RAPITS (Woods and Warren, 1995), whereas ITS-Engineering (Srisethanil and Baker, 1996) switches its teaching styles based on the student's p reference, the learning outcome required and the subject matter.
The use of an accurate student model can greatly enhance the tutoring process. It can be used to determine gaps in the users knowledge and note which teaching strategies the user prefers, or seems to understand better, and react accordingly. If a user is having problems learning with one strategy, the system can automatically change to an alternative strategy, especially one for which the user has previously shown a preference. Automatic selection of teaching strategies is the most desirable of the selection techniques that can be used. This is the way a human tutor reacts to a user when they are having trouble understanding some concept. Typically, the same information is presented again in a slightly different form. This allows alternative feedback to be generated about the same problem and hopefully, ease the learning process.
Unfortunately, the construction, use and maintenance of these student models is extremely difficult. Roberts (1993) maintains that the definition of the student model and the subsequent interpretation of user actions with respect to that model becomes a daunting problem. It is also a non trivial task for a student model to be verified as it is a model of some part of the users current understanding. Also, to make a model that is detailed enough to be useful requires a large amount of work. Muhlhauser (1990) observes that the enormous complexity of a human mind makes a good model hard to build and can lead to a lack of general applicability of the ITS developed.
Also, the development of a student model is not a finite process. It must be continually updated and maintained so that it can refine and adjust its
model in order to present a more accurate view of the user. This can be done dynamically through a tutoring session, which can increase the computational requirements of the system, or after a session, which requires recording the session and extensive post-session processing.
When a student model is being used in a tutoring session, the system will be continually comparing the model to the current actions of the user. This requires a comprehensive diagnostic and monitoring component for the system which may be difficult to keep independent from the domain material. This is particularly a problem when alternative types of user interfaces are to be developed, e.g. textual or graphical interfaces. In this case, providing a general purpose monitoring system may be impossible. If a student model is not to be used then alternative approaches for teaching strategy selection must be found. One approach is that of user selection. Before a tutoring session has commenced, either a tutor can select an appropriate strategy for a user (for example, based on previous performance) or a user may select his /her own strategy of preference. Once a selection has been made, this strategy is used throughout the remainder of the tutoring session. If the user wishes to change the strategy, the current session is exited, a new strategy selected and the session recommenced.
A major benefit of this approach is that the huge overhead of development and maintenance of a complex student model is removed from the tutoring system development process. Also, it allows the user to control how they learn and can encourage interactive tutor supervision between a human tutor, the system and the user. Another benefit is that when one teaching strategy is selected, the teaching method is focused for that session. This allows the user to concentrate on the task at hand and they do not have to contend with alternating teaching styles, which may be distracting.
There are two main disadvantages with this approach. Firstly, the tutoring system is not totally independent. Some user input is required before a teaching session can begin. This may only be undesirable in some domain or particular problem-solving areas where a totally independent system may be required. Secondly, once a session has been started, the teaching strategy cannot be changed until the session is terminated. A
possible solution to this would be to allow sets of similar strategies to be grouped and allow the user to interactively upgrade the current strategy to a different one in its strategy set. This may be necessary if different teaching strategies are associated with different types of user interfaces. For example, all the strategies involving a graphical interface may be grouped, as textual based strategies would not be appropriate for the same session and vice versa.