Pergamon Expert Systems WithApplications, Vol. 12, No. 2, pp. 225-245, 1997 © 1997 Elsevier Science Ltd Printed in Great Britain. All fights reserved 0957-4174/97 $17.00+0.00 PII: S0957-4174(96)00097-8 Knowledge-Based Support for User Adapted Interaction Design 1 D. AKOUMIANAKIS AND C. STEPHANIDIS'~ Institute of ComputerScience, Foundation for Researchand Technology-Hellas (FORTH),Scienceand TechnologyPark of Crete, EO. Box 1385 Heraklion, Crete, GR-71110,Greece Abstract--The design of high quality user interfaces is increasingly becoming a knowledge-intensive task. Designers of user interfaces require tools to: (i) provide 'assistance' towards optimal decision making in varying design situations, (ii) 'critique' the quality of tentative design alternatives, (iii) 'propagate' and "reuse' the accumulated design wisdom resulting from past experience, best practice and existing knowledge and (iv) 'automate' the design of certain aspects of the user-~computer interaction. This paper briefly reviews the architecture and scope of a knowledge-based user interface design aid, called USE-IT, and describes its knowledge base, the inference facilities and the data structures that are available for capturing and embedding design decisions into user interface implementation. USE-IT exhibits characteristic properties which cross the boundaries of the known categories of knowledge- based user interface design tools, such as user interface design assistants, user interface design critics or user interface design generators. The primary objective of USE-IT is to provide the user interface design team with a suit of tools allowing the consolidation of tentative designs, their re-use and incremental evolution, as well as the automatic generation of the lexical specification of a user interface. Such a specification can be subsequently consulted and interpreted by a user interface development toolkit so as to realize the specified lexical interactive behaviour on a target plaO~orm (e.g. MS-Windows). © 1997 Elsevier Science Ltd INTRODUCTION THE NEED FOR THE APPLICATIONOF KNOWLEDGEbased techniques and expert systems, during the user interface design process, is now acknowledged in the human- computer interaction (HCI) community (Nakakoji et al., 1996). In recent years, the value of this synergy has been realised in a number of contexts, such as intelligent user interface design (Chignell & Hancock, 1987; Elkerton & Williges, 1988; Myers, 1991), model-based user inter- face development (Foley et al., 1991; Neches et al., 1993), adaptive user interfaces (Dieterich et al., 1993), user modelling components and systems (Brajnik & Tasso, 1994; Kobsa, 1993) and unified interface speci- fication (Stephanidis, 1995). All these efforts are characterised by a conscious search for adequate repre- sentation formalisms to encapsulate the required design knowledge, as well as the development and/or applica- tion of reasoning mechanisms to address user interface specific problems. 1 An initial description of the USE-ITsystemreferredto in this paper was reported in the ConferenceProceedingsof the 3rd WorldCongress on Expert Systems,Seoul, Korea 1996. t Author for correspondence. In general, the user-computer interaction is realized at three different levels (Foley et al., 1984): (1) The semantic level, which concerns the general functionality which needs to be made accessible to the user (e.g. editing files, storing/retrieving records). (2) The syntactic level, which concerns the structure and the syntax of the dialogue with which the semantics are made accessible to the user (e.g. steps taken by the user, tasks to be accomplished and decomposi- tion of tasks). (3) The lexical level, sometimes called the physical level, which concerns the structure of the actual input/output items which are used to physically realise the dialogue (e.g. interaction objects like menus, list boxes, buttons as well as input devices, and interaction techniques); it should be mentioned that, physically, interaction is carried out at the lexical level. These three levels of designing interactions give rise to a large and complicated design space, whose seman- tics are difficult to encapsulate and consolidate into suitable models. Moreover, what is even less clear is how 225