notesum.ai

Published at October 21

STAR: A Simple Training-free Approach for Recommendations using Large Language Models

cs.LG

cs.AI

cs.RO

stat.ML

Released Date: October 21, 2024

Authors: Dong-Ho Lee¹, Adam Kraft², Long Jin³, Nikhil Mehta², Taibai Xu³, Lichan Hong², Ed H. Chi², Xinyang Yi²

Aff.: ¹University of Southern California; ²Google DeepMind; ³Google

Arxiv: https://arxiv.org/abs/2410.16458v1

Refer to caption

Category	Method / Model	Train	Beauty				Toys and Games				Sports and Outdoors
Category	Method / Model	Train	H@5	N@5	H@10	N@10	H@5	N@5	H@10	N@10	H@5	N@5	H@10	N@10
Baseline	KNN	✓	0.004	0.003	0.007	0.004	0.004	0.003	0.007	0.004	0.001	0.001	0.002	0.001
	Caser (tang2018personalized, )	✓	0.021	0.013	0.035	0.018	0.017	0.011	0.027	0.014	0.012	0.007	0.019	0.010
	HGN (ma2019hierarchical, )	✓	0.033	0.021	0.051	0.027	0.032	0.022	0.050	0.028	0.019	0.012	0.031	0.016
	GRU4Rec (hidasi2015session, )	✓	0.016	0.010	0.028	0.014	0.010	0.006	0.018	0.008	0.013	0.009	0.020	0.011
	BERT4Rec (sun2019bert4rec, )	✓	0.020	0.012	0.035	0.017	0.012	0.007	0.020	0.010	0.012	0.008	0.019	0.010
	FDSA (zhang2019feature, )	✓	0.027	0.016	0.041	0.021	0.023	0.014	0.038	0.019	0.018	0.012	0.029	0.016
	SASRec (kang2018self, )	✓	0.039	0.025	0.061	0.032	0.046	0.031	0.068	0.037	0.023	0.015	0.035	0.019
	$\bf{S^{3}}$ -Rec (zhou2020s3, )	✓	0.039	0.024	0.065	0.033	0.044	0.029	0.070	0.038	0.025	0.016	0.039	0.020
	P5 (zhou2020s3, )	✓	0.016	0.011	0.025	0.014	0.007	0.005	0.012	0.007	0.006	0.004	0.010	0.005
	TIGER (geng2022recommendation, )	✓	0.045	0.032	0.065	0.038	0.052	0.037	0.071	0.043	0.026	0.018	0.040	0.023
	IDGenRec (tan2024idgenrec, )	✓	0.062	0.049	0.081	0.054	0.066	0.048	0.087	0.055	0.043	0.033	0.057	0.037
STAR-Retrieval	-	✗	0.068	0.048	0.098	0.057	0.086	0.061	0.118	0.071	0.038	0.026	0.054	0.031
STAR-Ranking	point-wise	✗	0.068	0.047	0.096	0.056	0.086	0.061	0.117	0.071	0.037	0.026	0.054	0.031
	pair-wise	✗	0.072	0.051	0.101	0.060	0.090	0.064	0.120	0.073	0.040	0.028	0.056	0.034
	list-wise	✗	0.065	0.047	0.090	0.055	0.083	0.060	0.111	0.069	0.036	0.026	0.052	0.031