Rutgers' TREC-6 Interactive Track Experience

The goal of the Rutgers TREC-6 Interactive Track study was to compare the performance and usability of a system offering positive relevance feedback with one offering positive and negative relevance feedback. Our hypothesis was that the latter system would better support the aspect identification task than the former. Although aspectual recall was higher for the system supporting both kinds of relevance feedback (0.53 vs. 0.46), the difference was not significant, possibly because of the small number of subjects (four in each condition, each doing three searches). Usability results were also equivocal, perhaps due to the complexity of the system. Compared to ZPRISE, the control system without relevance feedback, both relevance feedback systems were rated more difficult to learn to use, but more effective.

• Type: Conference
• Year: 1997
• Where: TREC
• Authors: Nicholas J. Belkin, Jose Perez Carballo, Colleen Cool, Shin-jeng Lin, Soyeon Park, Soo Young Rieh, Pamela A. Savage-Knepshield, C. Sikora, H. Xie, James Allan
• Keywords: Negative Relevance Feedback, Positive Relevance Feedback, Relevance Feedback, TREC 1997, TREC 2008