Information Management Usage Models. When the computer is helping a user to organize and find relevant information, there will undoubtedly be errors. This work attempts to gauge the amount of trust that must exist between such a system and the user in order for the system to be useful. Additionally, this work attempts to discover user interface models and problem domains that will help mitigate the issue of trust. Another important issue is the amount of work that is introduced by classification errors. If the goal of the system is to reduce the information overload on the user, but the system classifies results with 5% error, then there is a possibility that the cost required by the user to correct the errors actually results in a decrease in productivity if the system were not used at all. To better understand these extremes, I have been comparing a manual classification system (such as existing Personal Information Managers, or PIMs) vs. an automatic classification system.
Arctic Ice Atlas. This project is driven by UAA Engineering and Geomatics. The goal is to create an online GIS-based Alaska Sea Ice Atlas so that government, industry, and the public can better perform risk assessment for ship navigation or other ice-related engineering projects. From the computer-science perspective, there are several challenges in providing a usable web-based interface to a massive database (many gigabytes) of ice observations. More info is available from the Alaska Sea Ice Homepage.
Application of Relevance Technology. Relevance technology constitutes the algorithms and associated interfaces that allow users to find things related to some other thing. This facilitates context and knowledge discovery. There are many places where this might be applied to business problems or elsewhere. One example is the application of relevance technology to documents such as email, so that related emails or files may be automatically invoked based upon the user's current context. Another example is relevance among people, so that you might find other people interested in something that you are interested in. One context this might be used is something like Instant Messaging -- to be able to find other people chatting about what you're chatting about. Similarly, online chatters could send messages to others via "Topic Chat." By building a profile of terms and topics that a person chats about, one's messages could automatically be forwarded to all other interested parties instead of subscribing to a particular channel or room. Relevance technology can also help in indexing, visualizing, and retrieving logs of chat information. Finally, another example is physical objects such as books, so that given a book, a user can easily find other books, comments, and other data related to the book.
Machine Learning of Personality Profiles. Previous work in the social sciences has demonstrated that computer agents with personalities may subtly affect user reactions. Users may unconsciously like or dislike the computer depending on its personality. This project investigates ways to automatically learn personalities based upon an analysis of user input. By determining knowledge structures important to personality, input can be parsed and mapped to the knowledge structures for future output. The current domain is an Internet Relay Chat gamebot that hosts trivia and word games. The goal is to augment the cyberspace hosts with personalities to make the game more enjoyable for the players.
Information Filtering. My dissertation from 1996 involved the filtering of Usenet news by learning user profiles.