|M.Sc Student||Zendel Oleg|
|Subject||Information Needs, Queries and Query Performance Prediction|
|Department||Department of Industrial Engineering and Management||Supervisor||PROF. Oren Kurland|
|Full Thesis text|
The query performance prediction (QPP) task is to estimate ad hoc retrieval effectiveness with no relevance judgments. The vast majority of previous work on QPP has not accounted for the effectiveness of a query in representing the underlying information need for retrieval.
We empirically show that this fact has far reaching implications: the relative prediction quality of various predictors changes when changing the effectiveness of the query used for retrieval.
We present a novel probabilistic framework for QPP which does account for the connection between the information need and the query used to represent it. In addition, we address a variant of the QPP task where in addition to the query for which prediction is performed, other queries that represent the information need are available. We devise a formal probabilistic framework to address this specific QPP challenge and empirically demonstrate the clear merits of predictors instantiated from this framework.
In the second part of this work, we address a novel prediction task: predicting topic difficulty or information need difficulty. That is, the goal is to estimate for a given information need, retrieval method and a corpus, the effectiveness of search performed for *some* query used to represent the need.
We present a probabilistic framework to this end and empirically demonstrate it merits with respect to some approaches to quantifying topic difficulty.