Standard Deviation as a Query Hardness Estimator

In this paper a new Query Performance Prediction method is introduced. This method is based on the hypothesis that different score distributions appear for ‘hard’ and ‘easy’ queries. Following we propose a set of measures which try to capture the differences between both types of distributions, focusing on the dispersion degree among the scores. We have applied some variants of the classic standard deviation and have studied methods to find out the most suitable size of the ranking list for these measures. Finally, we present the results obtained performing the experiments on two different data-sets.

• Type: Journal
• Year: 2010
• Where: SPIRE
• Authors: Joaquín Pérez-Iglesias, Lourdes Araujo
• Keywords: Classic Standard Deviation, Information Technology, Query Performance Prediction, Score Distributions, SPIRE 2010