Predicting IR personalization performance using pre-retrieval query predictors

Although personalization generally improves query performance, it may also occasionally harm how queries perform. If we are able to predict and therefore disable personalization for such situations, overall performance will be higher and users will be more satisfied with personalized systems. We use various state-of-the-art, pre-retrieval query performance predictors and propose several others including user profile information for this purpose. We study the correlations between these predictors and the difference between personalized and original queries. We also use classification and regression techniques to improve the results and finally achieve slightly more than one third of maximum ideal performance. We consider this to be a good starting point within this research line, which will undoubtedly result in further work and improvements.