An empirical model for tutoring strategy selection in multimedia tutoring systems

Abstract

This paper proposes an empirical model for tutoring strategy selection in multimedia tutoring systems based on factors that influence human tutoring strategy selection. It also demonstrates how this model is used in a multimedia tutoring system and assesses the benefits through the comparative evaluation of two multimedia tutoring systems, one that includes the model and one that does not.

Introduction

In empirical studies of human teaching in the 1990s, the use of different tutoring strategies has been found to form a substantial part of the overall teaching interaction. It is certainly important for a tutoring system to provide more than one tutoring strategy because if the system has only one tutoring strategy to use, it may not be able to adapt promptly to the changing cognitive needs of the student [21]. Researchers in tutoring systems are looking for the knowledge underpinning demonstrably effective ways of tutoring [18]. The pressing need is not just for knowledge on optimal ways of tutoring, but for knowledge about how to tutor in an effective way. As with overall tutoring, the same applies to the use of tutoring strategies. This paper argues that the focus of knowledge on the use of multiple tutoring strategies is on when, why and which one to use. pepe [31] is a planning framework that allows the designer of an intelligent tutoring system (ITS) to incorporate and to encode a variety of tutoring strategies in an ITS. The main focus of pepe, however, is on content planning as opposed to delivery planning. This separation refers to the distinction between the subject matter and the formats in which it can be presented. Content planning entails the generation, ordering and selection of content goals, and the monitoring of the execution of the content plan. Delivery planning involves choosing the actual activities and interactions that help the student achieve his goals. pepe focuses on the content issues alone. Although pepe sets out to accommodate the use of more than one tutoring strategies, it does not describe how the selection should proceed, nor does it describe how content planning can be used for tutoring strategy selection.

Eon [17] is a set of authoring tools for ITSs designed to be used by human tutors and instructional designers to represent their own tutoring strategies. The primary focus of Eon is on the use of multiple tutoring strategies. This “meta-strategic” knowledge base is distinct from the tutoring strategy knowledge base. Eon provides tools for describing the subject domain as network-related topics, creating reusable interactive presentation screens, and authoring the procedures, which tell the tutoring system how and when to interact with the student. In terms of tutoring strategy selection, Eon uses a parameterised approach that allows authors of ITSs to define a number of tutoring strategy parameters, e.g. ‘degree of hinting’ and ‘degree of interruption’. Eon currently offers a general mechanism for representing tutoring strategies, but does not, in itself, contain any specific tutoring strategies. The success of the tutoring strategies used has, therefore, not been taken into account. Tutoring strategies have subsequently not been linked with tutorial goals.

One of the most significant contributions in this area is the “dynamic method selection” in gte [30]. gte is a Generic Tutoring Environment for developing courseware, based upon a generic instructional task-based approach. The selection takes into account instructional tasks, instructional methods and instructional objects. Instructional tasks are activities accomplished by the tutor in a tutorial. Instructional methods are procedures to carry out tasks, which are taken as equivalent to tutoring strategies in the context of this paper. Instructional objects are means that methods employ. Instructional methods are chosen according to a numerical applicability value for each method. These values are calculated according to a list of conditions attached to the individual methods. While gte has comprehensively represented the instructional processes from the tutor's viewpoint, the learner's perspective, which is also important in the tutoring process, has not been sufficiently addressed [6]. gte's selection could be improved if it could couple a student's learning results with the selection mechanism. There is no separate knowledge base for the selection procedure in gte. While having such knowledge embedded in the tutoring strategies themselves makes their context more explicit, it obscures the relationship between the various elements within the selection mechanism, thus lessening the transparency between the different aspects, which affect selection [17].

These examples provide evidence that although many systems use more than one tutoring strategy and that they all incorporate some kind of support for tutoring strategy selection, they lack a common basis on which the rules governing the selection can be applied. The lack of common basis for tutoring strategy selection supports the fact that what is missing is a formalisation of such a process. The benefits of formalising tutoring strategy selection in tutoring systems are to ensure that tutoring systems consider each tutoring strategy according to [7], [12], [13], [22]: what the system aims to teach; the individual student; the particular situation or application the tutoring strategy is to be used for; the previous failures and successes of a tutoring strategy in similar situations; what a human tutor would have done; and to ensure in general that tutoring strategy selection is provided for in an adaptive and yet coherent manner.

This paper does not propose new research in computational tutoring strategies. Its purpose is to model the process of tutoring strategy selection in multimedia tutoring systems. Consequently, the paper is organised as follows. First, it constructs a model for tutoring strategy selection based on factors that influence human tutoring strategy selection, followed by a comparative evaluation of sonata and aristotle, two multimedia tutoring systems that use multiple tutoring strategies, but only aristotle incorporates the model for tutoring strategy selection in its architecture.