AT&T at TREC-9

This year we come to TREC with a new retrieval system Tivra that we have implemented over thelast year. Tivra is based on the vector space model, and is mainly designed to do large-scale web searchwith limited resources. We run Tivra on a cheap Linux box. It currently indexes around 14-15 gigabytesof web data per hour, and allows sub-second web searches for 2-3 word queries on a 700 MHz Pentiumbox. At the time of submissions Tivra was in its early development stages, and was not fully...