notesum.ai

Published at December 4

Beyond Questions: Leveraging ColBERT for Keyphrase Search

cs.IR

Released Date: December 4, 2024

Authors: Jorge Gabín¹, Javier Parapar², Craig Macdonald³

Aff.: ¹Linknovate Science, Santiago de Compostela, Spain; ²IRLab, CITIC, Computer Science Department, University of A Coruña, A Coruña, Spain; ³University of Glasgow, Glasgow, Scotland

Arxiv: http://arxiv.org/pdf/2412.03193v1

Model	TREC DL 2019			TREC DL 2020			MSMarco Dev Small
Model	MAP@1k	nDCG@10	MRR@10	MAP@1k	nDCG@10	MRR@10	MRR@10	R@50	R@200	R@1k
BM25	0.3031	0.4989	0.6780	0.2974	0.4793	0.6457	0.1850	0.5952	0.7481	0.8677
BM25 >> ColBERT	0.4596	0.7039	0.8353	0.4828	0.7146	0.8380	0.3524	0.7909	0.8523	0.8677
BM25 >> ColBERTKP_QD	0.4608	0.7107^⋄	0.8469^⋄	0.4797^⋄	0.7096^⋄	0.8627 ^⋄	0.3507	0.7871	0.8524	0.8677
BM25 >> ColBERTKP_Q & ColBERT_D	0.4579^⋄	0.7062^⋄	0.8553^⋄	0.4816^⋄	0.7154 ^⋄	0.8356^⋄	0.3556	0.7841	0.8531	0.8677
ColBERT	0.4452	0.7078	0.8574	0.4692	0.6866	0.8318	0.3572	0.8221	0.9088	0.9491
ColBERTKP_QD	0.4378^⋏	0.7140 ^⋏	0.8667 ^⋏	0.4609^⋏	0.6892^⋏	0.8576^⋏	0.3541	0.8179	0.9081	0.9495
ColBERTKP_Q & ColBERT_D	0.4350^⋏	0.7075^⋏	0.8566^⋏	0.4580^⋏	0.6798^⋏	0.8602^⋏	0.3527	0.8116^⋏	0.9040	0.9462