Abstract

Experiments relating to TREC-7 Ad Hoc, HP and VLC tasks are described and results reported. Minor remnants of last year’s Ad Hoc methods do not appear to have resulted in worthwhile improvements in performance. However, larger benefit were gained from automatic feedback than last year and concept scoring was very beneficial in the Manual AdHoc category. In the Automatic Ad Hoc category title-only performance seems to have suffered more severely than long-topic from an umber of lexical scanning shortcomings and from an excessive stop word list. The HP track was used to validate the usability of the combination of PADRE and the Quokk a GUI. In the VLC track, the 100 gigabyte collection was indexed in under eight hours and moderately effective queries were processed in less than two seconds.
‚Äč