| The purpose of Knowledge Discovery and Data-mining (KDD) systems is to uncover novel, non-trivial patterns; however, these systems also discover many useless patterns. The interpretation and evaluation of patterns requires domain expertise. The process is inherently subjective because experts differ in experience and knowledge. In most KDD research, interpretation/evaluation has been done by the researchers or by a single expert. In this research, we introduce Collaborative Interpreting (CI), which uses the collective expertise of a diverse group of experts to evaluate data-mining results. Our hypothesis is that the collaborative interpretation of the group would be better than that of any single expert1.; This research involves the development of a World Wide Web (WWW)-based Collaborative Interpreting System (CIS), which employs an iterative group decision-making process based on the Delphi Method. The CIS compiled the interpretations of a panel of geographically and temporally distributed experts on a continuous basis. The application domain used was hospital infection surveillance. The panel consisted of two infectious disease physicians, two critical care physicians, a pharmacist, and a clinical microbiologist. Patterns were generated for five months by the Data Mining Surveillance System (DMSS)2 using retrospective clinical laboratory infection-control data from the University of Alabama at Birmingham Hospital. For each month, the panel used CIS to judge the importance of the patterns independently and then collectively with regard to epidemiology investigation priorities.; Group and individual judgments were compared to those of a criterion judge who is an infectious disease physician with more than thirty years of infection control experience. In addition to the patterns, he was provided with patient chart review data. The results showed that the panel's collective judgments were generally in greater agreement with the criterion than any individual member of the panel. The results confirmed the hypothesis and showed that interpretation is a more complex problem than generally assumed.; The findings are encouraging that CIS can (1) utilize collective expertise to enhance the interpretation of KDD results, (2) enrich the knowledge base of experts through collaboration with other experts, and (3) potentially help refine data-mining algorithms to reduce generation of trivial patterns.; 1Turoff M: Computer-Mediated Communication Requirements for Group Support. Journal of Organizational Computing 1991; 1: 85–113. 2DMSS was originally developed at the University of Alabama at Birmingham by S. Brossette. |
