Juru at TREC 2004: Experiments with Prediction of Query Difficulty

Our experiments in the Robust track this year focused on predicting query diculty and using this prediction for improving information retrieval. We developed two prediction algorithms and used the subsequent prediction in several ways in order to improve the performance of the search engine. These included modifying the search engine parameters, using selective query expansion, and switching between dierent topic parts. We also experimented with a new scoring model based on ideas from the field of machine learning. Our results show that query prediction is indeed ecient in improving retrieval, although further work is needed in order to improve the performance of the prediction algorithms and their uses.