Knowledge-based assistance for accessing large, poorly structured information spaces
Frank Curtis Stevens · 1993
Large information spaces present several problems to people searching for interesting information including information overload. This thesis describes a novel approach to handling overload problems in the domain of Usenet News, an open access computer-based bulletin board system that distributes messages and software. A conceptual framework is developed that shows the need for (a) flexible organization of information access interfaces, (b) personalized structure to deal with vocabulary mismatches and individual information needs, and (c) semi-autonomous agents that assist in creating this personalized structure. In addition, an operational system (scINFOSCOPE) instantiate this framework allowing for the exploration and evaluation of the approach in realistic working environments. Using scINFOSCOPE, Usenet readers evolve the predefined system structure to suit their own tasks or semantic interpretations. Agents assist users by suggesting filters based on observed interest patterns. Evaluation of the system indicates that (1) personalized structure is quickly adopted and effectively used; (2) personalized structure reduces information overload caused by uninteresting messages; and (3) scINFOSCOPE agents are an effective aid to the creation of personalized structure in Usenet News. However, user studies have exposed deficiencies in the current implementation that indicate a more sophisticated suite of agents could increase the usefulness of many suggestions. (Also available from the University of Colorado at Boulder Computer Science Department as Technical Report # CU-CS-640-93.)