notesum.ai

Published at November 18

QARM: Quantitative Alignment Multi-Modal Recommendation at Kuaishou

cs.IR

cs.AI

Released Date: November 18, 2024

Authors: Xinchen Luo¹, Jiangxia Cao¹, Tianyu Sun¹, Jinkai Yu¹, Rui Huang¹, Wei Yuan¹, Hezheng Lin¹, Yichen Zheng¹, Shiyao Wang¹, Qigen Hu¹, Changqing Qiu¹, Jiaqi Zhang¹, Xu Zhang¹, Zhiheng Yan¹, Jingming Zhang¹, Simin Zhang¹, Mingxing Wen¹, Zhaojie Liu¹, Kun Gai¹, Guorui Zhou¹

Aff.: ¹Kuaishou Technology, Beijing, China

Arxiv: http://arxiv.org/abs/2411.11739v1

QARM Model Variants	Shopping
	CTR			CVR			CTCVR
	AUC	UAUC	GAUC	AUC	UAUC	GAUC	AUC	UAUC	GAUC
Baseline Model	79.31%	66.48%	68.05%	86.79%	69.13%	69.82%	87.92%	70.70%	71.41%
+ IA Rep	-0.03%	+0.02%	-0.01%	-0.02%	+0.02%	+0.03%	-0.07%	+0.13%	+0.020%
+ VQ Code	+0.09%	+0.20%	+0.21%	+0.11%	+0.38%	+0.38%	+0.19%	+0.49%	+0.52%
+ RQ Code	+0.12%	+0.33%	+0.26%	+0.14%	+0.32%	+0.34%	+0.18%	+0.50%	+0.40%
+ VQ & RQ Code	+0.23%	+0.56%	+0.50%	+0.26%	+0.77%	+0.77%	+0.35%	+1.00%	+0.92%