notesum.ai

Published at November 18

TSPRank: Bridging Pairwise and Listwise Methods with a Bilinear Travelling Salesman Model

cs.AI

cs.IR

Released Date: November 18, 2024

Authors: Weixian Waylon Li¹, Yftah Ziser², Yifei Xie¹, Shay B. Cohen¹, Tiejun Ma¹

Aff.: ¹University of Edinburgh, United Kingdom; ²Nvidia Research, Israel

Arxiv: http://arxiv.org/abs/2411.12064v1

Refer to caption

		Top 10					Top 30
Model	Type	NDCG@3	NDCG@5	NDCG@10	MRR	$\tau$	NDCG@3	NDCG@5	NDCG@10	MRR	$\tau$
LambdaMART	Pairwise	0.6833	0.7222	0.8707	0.4259	0.1474	0.7340	0.7298	0.7403	0.3617	0.2372
Rankformer	Listwise	0.7220	0.7565	0.8865	0.4661	0.2317	0.7486	0.7470	0.7596	0.3732	0.2834
TSPRank-Local	Pairwise-Listwise	0.6858	0.7213	0.8719	0.4266	0.1544	0.7189	0.7240	0.7362	0.3206	0.2054
TSPRank-Global	Pairwise-Listwise	0.7281	0.7585	0.8884	0.4861	0.2212	0.7582	0.7558	0.7631	0.3895	0.2647