Overview of TREC-7 Very Large Collection Track

In line with the wishes of last year's participants, this year's VLC track was essentially a re-run of last year's with a ve-fold increase in data size. The data used was a completely new 100-gigabyte collection of Web documents (the VLC2) whose characteristics are presented here. This time, two orders of magnitude scale-up was investigated using 1%, and 10 % samples as well as the full collection. Six groups managed to complete the full VLC task, of which ve completed last year's track. An overview is given of the track participants, the methods used and the results obtained. One group of participants, using hardware costing less than $US10,000, have shown that a hundred gigabyte collection can be indexed in less than ten hours and that quite good rankings (better than several well-known search engines) can be produced from queries processed in less than one second