Improving robustness using query expansion

Abstract

This paper describes our work at CLEF 2007 Robust Task. We have applied local query expansion using windows of terms, but considering different measures of robustness during the training phase in order to optimize the performance: MAP, GMAP, MMR, GS@10, P@10, number of failed topics, number of topics below 0.1 MAP, and number of topics with P@10=0. The results were not disappointing, but no settings were found that simultaneously improved all measures. A key issue for us was to decide which set of measures we had to select for optimization. This year all our runs also gave good rankings, both base runs and expanded ones. However, our expansion technique does not improve significantly the retrieval performance. At TREC and CLEF Robust Tasks other expansion techniques have been used to improve robustness, but results were not uniform. In conclusion, regarding robustness the objective must be to make good information retrieval systems, rather than to tune some query expansion techniques. © 2008 Springer-Verlag Berlin Heidelberg.