Navigating the User Query Space

Query performance prediction (QPP) aims to automatically estimate the performance of a query. Recently there have been many attempts to use these predictors to estimate whether a perturbed version of a query will outperform the original version. In essence, these approaches attempt to navigate the space of queries in a guided manner. In this paper, we perform an analysis of the query space over a substantial number of queries and show that (1) users tend to be able to extract queries that perform in the top 5% of all possible user queries for a specific topic, (2) that post-retrieval predictors outperform preretrieval predictors at the high end of the query space. And, finally (3), we show that some post retrieval predictors are better able to select high performing queries from a group of user queries for the same topic.