Tailoring the Presentation of Plans to Users’ Knowledge and Capabilities Detlef Küpper Dept. of Mathematics and Computer Science University of Essen D-45117 Essen, Germany +49 7732 54872 Detlef.Kuepper@epost.de Alfred Kobsa Dept. of Information and Computer Science University of California Irvine, CA 92697-3425, U.S.A. +1 949 824-3007 kobsa@ics.uci.edu ABSTRACT Tailoring advice to a user means finding a plan by which she can reach her goal, and supplying the missing knowl- edge that she needs to successfully execute the plan. The paper presents an algorithm to determine the kind and amount of this missing knowledge for an already generated domain plan. We show that both the user's knowledge and his capabilities to perform actions must be taken into account when deciding on a plan presentation that is ade- quate for him. We also demonstrate that it may be useful to consider issues of plan presentation already during the planning process, and indicate how this can be accom- plished in a planning system. Keywords User-tailored advice, plan presentation, user modeling INTRODUCTION The effectiveness of advice-giving systems essentially hinges on users’ ability to take advantage of the given advice to reach their goals. The problem of whether a user can understand an advice already received considerable attention in the research literature (see, e.g. [22] for a sur- vey). However, an additional prerequisite for the success of advice is that the user must be able to execute the advice, i.e. have the capabilities to perform each step of the advice. Capability in this sense means the user’s personal abilities and her authorization to perform the actions that occur in the advice. In [11], we proposed a plan generation approach to achieve the user’s goals that considers the user’s capabili- ties. The resulting plans are therefore in principle executa- ble by the user. In order to perform the plan, she may however still need additional information. This possibly missing knowledge – or in general the gap between a user’s capabilities to perform plans and his knowledge how to do this – determines the scope of user-tailored advice. In this paper, we describe our approach to identify the knowledge that the current user still needs to perform a plan. We start with a plan that the user can in principle execute and that is suitable for reaching the user’s goals. We will show that the knowledge that is still required depends not only on the user’s knowledge but also her capabilities. The rest of the paper is organized as follows. First, we summarize central characteristics of the user model that we employ and introduce some terminology. The next section discusses the different types of knowledge that users need in order to perform a plan. We then present an algorithm that determines such knowledge for an already generated plan specifically for the current user. This algo- rithm also determines the “presentation cost” of the knowledge/plan to be communicated, which is propor- tional to the structural complexity of the presentation and can be regarded as a coarse estimate of the user’s compre- hension efforts. Afterwards, we re-integrate our results into the plan generation process to ascertain that plans will preferably be generated that the user is not only capable of performing, but that also have the lowest presentation costs. We also discuss other measures besides presentation cost for rating plan presentations that take user skills, user preferences, and the likelihood of success into account. Finally we describe some related research, and outline future work to decide which parts of the missing knowl- edge identified by our algorithm should be presented explicitly and which parts may omitted, depending on (assumed) inferences that the user may draw. ELEMENTS OF OUR APPROACH Our approach exploits an extension of the user model that was presented in [11]. This model separates the system’s assumptions about the user’s knowledge from its assump- tions about his capabilities. Capabilities of a user are mod- eled as plan operators that the user is in principle able and authorized to execute. These plan operators have a termi- nological representation in the user model, the so-called plan concepts. Preconditions and effects of plan operators are represented by attributes of their plan concepts. For the generation of user-tailored plans, plan operators/concepts can be instantiated and become the steps of the plans. Submitted to the 2002 Conference on Intelligent User Interfaces, San Francisco, CA