Adaptive Hypermedia for Support Systems Gernoth Grunst German National Research Center for C.S. presented by: David Mark Carlson System: HyPLAN Goal: Adaptive help for non-trivial computer application. What is HyPLAN? Domain: Excel (Nontrivial - heterogeneous, open-ended). - Help system does not change user's view of Excel (ExcelInExcel). - Provides task sensitive help (as recognized by the PLANET system). - Help is presented in hypertext format (HyTask). ExcelInExcel - Layer above Excel, captures user actions and passes them to loging mechanism as well as to Excel. - Semi-interesting hack (built on macro system of Excel so it is not general). PLANET Goal: Recognize what the user is trying to do. - Events from ExcelInExcel are recorded and entered into trees. - Experts have generated trees which represent tasks. - PLANET does pattern recognition to determine top 5 plans and uses these as the entrypoint into HyTASK. figure 4 HyTASK - Stand-alone hypertext help facility. - Developed in Hypercard, moved to MediaTracks. - Immediate focus is on "how to do" animations. - Supplemented with "how it works" illustrations. figure 5 Why is this paper interesting? - System uses plan recognition. Been there. - Help is presented using hypertext. Done that. Interesting facet of this paper is the focus on usability and use of HCI methodology. Methodology: Early focus on users, iterative design, rapid prototyping. (Think Aloud Protocols) User Population: Casual users who have learned the basics, but forgotten most everything. Early Prototype: Human expert provides help. "Human model is the most advanced example of cognitive estimation competence." Results: - (semi) clear goals. - Ability to start work before AI successfully mimics human intelligence. BIG Result: Allows focus on how Adaptive systems should be before they are built. Learned (among other things): How to do hypertext. reduce text, animation must be good. How to recognize plans. significant functions, inaccuracy, utilization of error messages. How to intervene: Don't. Users are annoyed by human intervention, machine intervention could not be significantly better. Users must self-asses.