We propose QLFusion, an approach based on Quantification Learning (QL) to improve rank fusion performance in Information Retrieval. We first introduce a QL model based on a Bayesian Neural Network to estimate the proportion of relevant documents in a ranked list. The proposed model is trained using a probabilistic loss function formulated specifically for this QL task. Next, we describe a rank fusion algorithm which leverages on this information to merge multiple ranked lists. We compare our approach to various popular rank fusion baselines on multiple collections, showing how the proposed approach outperforms the baselines in several evaluation measures.

A Bayesian neural model for documents' relevance estimation

Purpura A.;Susto G. A.
2021

Abstract

We propose QLFusion, an approach based on Quantification Learning (QL) to improve rank fusion performance in Information Retrieval. We first introduce a QL model based on a Bayesian Neural Network to estimate the proportion of relevant documents in a ranked list. The proposed model is trained using a probabilistic loss function formulated specifically for this QL task. Next, we describe a rank fusion algorithm which leverages on this information to merge multiple ranked lists. We compare our approach to various popular rank fusion baselines on multiple collections, showing how the proposed approach outperforms the baselines in several evaluation measures.
2021
CEUR Workshop Proceedings
2nd International Conference on Design of Experimental Search and Information REtrieval Systems, DESIRES 2021
File in questo prodotto:
Non ci sono file associati a questo prodotto.
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11577/3402961
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
  • OpenAlex ND
social impact