notesum.ai

Published at November 22

IRLab@iKAT24: Learned Sparse Retrieval with Multi-aspect LLM Query Generation for Conversational Search

cs.IR

cs.CL

Released Date: November 22, 2024

Authors: Simon Lupart¹, Zahra Abbasiantaeb¹, Mohammad Aliannejadi¹

Aff.: ¹University of Amsterdam

Arxiv: http://arxiv.org/abs/2411.14739v1

Refer to caption

Runs	LLM	Retrieval	Reranker	nDCG@5	nDCG	MRR	Recall@100	P@20	mAP
(1) MQ4CS-QR-deberta	GPT-4	SPLADE	DebertaV3	0.5213	0.6098	0.8479	0.4366	0.5927	0.3486
(2) MQ4CS-QR-ensemble	GPT-4	SPLADE	Ensemble	0.5155	0.6192	0.8644	0.4657	0.6184	0.3698
(3) GPT4QR-deberta	GPT-4	SPLADE	DebertaV3	0.5071	0.5420	0.8315	0.4135	0.5684	0.3166
(4) GPT4QR-bm25-QD1	GPT-4	BM25	MiniLM	0.4768	0.4373	0.8476	0.3421	0.5199	0.2315
(5) HumanQR-deberta	Human	SPLADE	DebertaV3	0.4767	0.5496	0.7808	0.4263	0.5383	0.3072
(6) HumanQR-ensemble	Human	SPLADE	Ensemble	0.5384	0.5819	0.8483	0.4777	0.5922	0.3533