The paper entitled “When Reducing Representations Improves Performance”, by Andrea Pasin, Guglielmo Faggioli and Nicola Ferro (University of Padua), Raffaele Perego (Cnr-Isti) and Nicola Tonellotto (University of Pisa), won the Best Paper Honorable Mention at the 48th European Conference on Information Retrieval (ECIR 2026), Delft, The Netherlands, 29 March - 2 April.

The paper shows that in neural Information Retrieval, less can be more. By intelligently selecting subsets of embedding dimensions, we can significantly improve ranking effectiveness. We can also identify dimension subsets that generalize across queries and datasets, pointing toward more efficient and effective dense retrieval (read more).