Query Performance Prediction and Classification for Information Search Systems

Zhongmin Zhang, Jiawei Chen, Shengli Wu

Abstract: Automatic performance prediction and classification for information search results is useful in different scenarios. In this paper, we propose two score-based post-retrieval performance prediction methods. Both of them take magnitude and variance of resultant document scores into consideration at the same time. We also try to classify queries into three different classes: easy, medium, and hard by using a support vector machine-based approach. The experimental results show that the proposed predictors in this paper are very competitive compared with other predictors in the same category, and the support vector machine-based approach is effective for query classification.