notesum.ai

Published at November 21

Teaching MLPs to Master Heterogeneous Graph-Structured Knowledge for Efficient and Accurate Inference

Released Date: November 21, 2024

Authors: Yunhui Liu, Xinyi Gao, Tieke He, Jianhua Zhao, Hongzhi Yin

Arxiv: http://arxiv.org/abs/2411.14035v1

Refer to caption

Dataset	Eval	MLP	RSAGE	HG2M	$\Delta_{MLP}$	$\Delta_{HGNN}$	HG2M+	$\Delta_{MLP}$	$\Delta_{HGNN}$	$\Delta_{HG2M}$
TMDB	prod	72.17±0.28	81.97±0.55	79.58±0.52	7.41	-2.39	81.11±0.68	8.94	-0.86	1.53
	ind	70.68±2.85	80.09±2.28	71.60±2.03	0.92	-8.49	72.24±2.44	1.56	-7.85	0.64
	tran	72.54±0.51	82.44±0.37	81.57±0.46	9.03	-0.87	83.33±0.37	10.79	0.89	1.76
CroVal	prod	82.45±0.12	86.32±0.15	85.55±0.19	3.10	-0.77	88.02±0.16	5.57	1.70	2.47
	ind	82.36±0.68	86.06±0.76	83.14±0.29	0.78	-2.92	83.76±0.44	1.40	-2.30	0.62
	tran	82.47±0.08	86.39±0.24	86.15±0.24	3.68	-0.24	89.09±0.22	6.62	2.70	2.94
ArXiv	prod	64.25±0.16	70.76±0.16	68.91±0.17	4.66	-1.85	70.34±0.30	6.09	-0.42	1.43
	ind	64.02±0.97	77.55±1.00	65.81±0.48	1.79	-11.74	65.95±0.77	1.93	-11.60	0.14
	tran	64.30±0.25	69.07±0.21	69.69±0.30	5.39	0.62	71.44±0.29	7.14	2.37	1.75
IGB-549K-19	prod	55.83±0.16	67.42±0.41	67.53±0.18	11.70	0.11	69.15±0.31	13.32	1.73	1.62
	ind	55.19±1.94	57.88±1.56	58.16±1.63	2.97	0.28	57.85±1.64	2.66	-0.03	-0.31
	tran	55.99±0.32	69.81±0.24	69.88±0.25	13.89	0.07	71.97±0.45	15.98	2.16	2.09
IGB-549K-2K	prod	53.34±0.13	59.53±0.41	58.99±0.63	5.65	-0.54	60.79±0.69	7.45	1.26	1.80
	ind	52.84±1.79	54.25±1.48	57.16±1.63	4.32	2.91	57.65±2.17	4.81	3.40	0.49
	tran	53.47±0.56	60.86±0.16	59.44±0.42	5.97	-1.42	61.58±0.47	8.11	0.72	2.14
IGB-3M-19	prod	62.30±0.08	69.02±0.19	67.21±0.21	4.91	-1.81	67.62±0.29	5.32	-1.40	0.41
	ind	61.83±1.14	65.12±1.08	64.11±1.02	2.28	-1.01	64.04±1.23	2.21	-1.08	-0.07
	tran	62.42±0.35	70.00±0.26	67.99±0.13	5.57	-2.01	68.52±0.08	6.10	-1.48	0.53
Avg. Rank/Avg.		4.0	1.7	2.8	6.24	-1.21	1.5	7.78	0.34	1.54