notesum.ai

Published at November 29

ContextGNN: Beyond Two-Tower Recommendation Systems

cs.IR

cs.LG

Released Date: November 29, 2024

Authors: Yiwen Yuan¹, Zecheng Zhang¹, Xinwei He¹, Akihiro Nitta¹, Weihua Hu¹, Dong Wang¹, Manan Shah¹, Shenyang Huang¹, Blaž Stojanovič, Alan Krumholz¹, Jan Eric Lenssen¹, Jure Leskovec¹, Matthias Fey¹

Aff.: ¹Kumo.AI

Arxiv: http://arxiv.org/pdf/2411.19513v1

Task	Split	Light	Multi	Graph	NGCF	NBFNet	Shallow	Context
Task	Split	GBM	VAE	SAGE	NGCF	NBFNet	Item	GNN
rel-amazon
user-item-purchase	val	0.18	0.37	1.53	1.16	2.60	1.45	2.71
user-item-purchase	test	0.16	0.23	0.74	0.88	2.06	0.56	2.93
user-item-rate	val	0.22	0.34	1.42	1.22	1.61	1.62	2.68
user-item-rate	test	0.17	0.24	0.87	0.86	1.24	0.74	2.25
user-item-review	val	0.14	0.19	1.03	0.79	1.86	1.06	1.95
user-item-review	test	0.09	0.10	0.47	0.55	1.57	0.40	1.63
rel-hm
user-item-purchase	val	0.44	0.33	0.92	0.81	2.64	0.49	3.00
user-item-purchase	test	0.38	0.28	0.80	0.75	2.81	0.40	2.93
rel-stack
user-post-comment	val	0.04	0.07	0.43	0.52	15.17	0.05	15.22
user-post-comment	test	0.04	0.01	0.11	0.13	12.72	0.03	13.34
rel-trial
condition-	val	4.88	2.59	3.12	3.71	11.33	0.88	11.59
sponsor-run	test	4.82	2.47	2.89	3.88	11.36	0.85	11.65
site-sponsor-run	val	10.92	7.16	14.09	9.68	17.47	12.59	28.55
site-sponsor-run	test	8.40	6.17	10.70	6.54	19.06	10.66	28.02
Average ( $\uparrow$ )	val	2.30	1.46	2.82	2.24	7.56	2.35	9.22
Average ( $\uparrow$ )	test	2.01	1.29	2.08	1.72	7.71	1.81	9.23