Notes for class on 10/14/94 by Tonya Beers Papers: (1) Pragmatic User Modelling for Adaptive Interfaces (2) Lies, Damned Lies and Stereotypes: Pragmatic Approximations of Users both by Judy Kay Overheads: GOALS OF ADAPTIVE INTERFACES - Improve efficiency and effectiveness - Improve usability of complex systems - Enhance ease of use - Increase speed of system - Display what the user wants to see WHY MODEL USERS IN ADAPTIVE INTERFACES? To support: - Different tasks - Various goals - Changing needs - The level of user knowledge BENEFITS OF USER MODELS IN SYSTEM ADAPTIVITY - Advisory, consultant, and help systems - interprets request and tailors answer - eg: Unix Consultant - Filtering systems - assess what the user needs to know - eg: Grundy: recommends books user would enjoy - Tailoring output - individualizes information presentation - eg: Tailor: describes complex devices at appropriate level - Tailoring interaction and modality - alters user interaction - eg: Checs: chemical engineering CAD system; only "legal" actions are available - Intelligence teaching system - selects specific teaching goals - based on user's assessed or expressed needs DEFINITION OF USER MODELING TERMS - 'User Model'- explicit representation of user - stored as a single category - does not depict all features - 'Consumer'- a program that adapts its interaction based on user model - 'Components'- the basic units depicting the user COMMON ROLES OF THE USER MODEL IN ADAPTIVE SYSTEMS (PART ONE) - Interpret user actions 1) User inputs 2) Dialogue history - Generate system actions 1) Logical level 2) Physical level COMMON ROLES OF THE USER MODEL IN ADAPTIVE SYSTEMS (PART TWO) - Attempts to cooperate with the user HIGH<-------------------------->LOW Help System ITS - Attempts to alter the user HIGH<--------------------------->LOW ITS Help System Filtering TASKS INVOLVED IN USER MODELLING - Acquisition of the User Model 1) Information supplied by user or his agent 2) Observations 3) Told-about - Representation of User Model 1) Use of scalars 2) Hierarchical representation 3) Stereotypes and prerequisite knowledge CHALLENGES OF USER REPRESENTATION - Noisy data - User change - Knowledge can be difficult to assess - Possibility dialogue was influenced by others - How detailed and flexible will user model be? Discussion Questions: 1) Can you develop "light weight" user models and tools and address all four of the roles of a user model? A light weight model would not be as flexible; it wouldn't fit users as closely. It would also allow less adaptivity, but it might be a good compromise. 2) Do you think automatic adaptivity will work for a user who has no interest or knowledge in customizing his interface? Problems: (1) How does the system know what the user's interests are? (2) There may be user resistance. If users don't want to customize things themselves, why do it for them? It will most likely only be confusing for the user. A lot of mental effort may be required to adapt the mental model to the changing interface. Also, users are generally very knowledgeable about costs associated with various actions; based on this information, they form decisions regarding what is right for them, and avoid those actions which will slow them down, etc. We must keep in mind that even novice users want some control. An example that was discussed was reading news. When entering news, we are prompted to respond to queries regarding new news groups. We don't want to deal with this every time we want to read news. It might be good to provide options regarding when and what information is presented. This gives the user more control. 3) When, what aspect, and how much of the user model should be accessed? Is this too much responsibility for the average user? Any part of the system should be accessible. Then, if things become too complex for the user, he/she can go back to the previous state of the system and make appropriate changes. In general, the user should determine the level of responsibility and adaptiveness. The user model would simply be used to guide the interaction with the system. It would also be good for the user to be aware of the effects of different levels of adaptation. A good solution might be for the system to prompt the user when it wants to change. The user can then decide to accept or override the system recommendation. The user may, however, indicate a desire to have the system recommendation implemented at a more convenient time (rather than when in the middle of some process). Again, the assumptions and effects of the adaptive change should be available to the user. 4) A single user will fit multiple stereotypes. How do you decide which stereotype will apply? The various stereotypical categories could be combined to form a "new" stereotype. This however, brings up the question of how conflicts would be handled. In this case, the best solution would probably be to have the user decide. ` 5) By stereotyping a user and then mapping across domains, say from movies to books, aren't you assuming too much consistency in a person's behavior? If so, won't this degrade adaptivity? YES. 6) If consumer information is filtered by stereotyped models of a group, will the individual user adapt to the preferences of the group? It may be possible to adapt in both areas: the system and the user. Most likely, we will find central tendency of the group. The user model is similar to marketing demographics.