Stephen Tomlinson. In E. M. Voorhees and D. K. Harman, editors, Proceedings of the Tenth Text REtrieval Conference (TREC 2001). Gaithersburg, Maryland, November 2001. NIST Special Publication 500-250.
Hummingbird submitted ranked result sets for the topic relevance task of the TREC 2001 Web Track (10GB of web data) and for the monolingual Arabic task of the TREC 2001 Cross-Language Track (869MB of Arabic news data). SearchServer's Intuitive Searching™ tied or exceeded the median Precision@10 score in 46 of the 50 web queries. For the web queries, enabling SearchServer's document length normalization increased Precision@10 by 65% and average precision by 55%. SearchServer's option to square the importance of inverse document frequency (V2:4 vs. V2:3) increased Precision@10 by 8% and average precision by 12%. SearchServer’s stemming increased Precision@10 by 5% and average precision by 13%. For the Arabic queries, a combination of experimental Arabic morphological normalizations, Arabic stop words and pseudo-relevance feedback increased average precision by 53% and Precision@10 by 9%.
Last Updated: 2003 Feb 25
Comments are welcome at comments@stephent.com.