notesum.ai

Published at November 11

Subgraph Retrieval Enhanced by Graph-Text Alignment for Commonsense Question Answering

cs.LG

cs.AI

cs.CL

cs.SI

Released Date: November 11, 2024

Authors: Boci Peng¹, Yongchao Liu², Xiaohe Bo³, Sheng Tian², Baokun Wang², Chuntao Hong², Yan Zhang¹

Aff.: ¹School of Intelligence Science and Technology, Peking University, China; ²Ant Group, China; ³School of Artificial Intelligence, Beijing Normal University, China

Arxiv: http://arxiv.org/abs/2411.06866v1

Methods	CommonsenseQA		OpenBookQA
Methods	IHdev-Acc (%)	IHtest-Acc (%)	RoBERTa-Large (%)	AristoRoBERTa (%)
Fine-tuned LMs	73.07 (±0.45)	68.69 (±0.56)	64.80 (±2.37)	78.40 (±1.64)
+ RN	74.57 (±0.91)	69.08 (±0.21)	65.20 (±1.18)	75.35 (±1.39)
+ RGCN	72.69 (±0.19)	68.41 (±0.66)	62.45 (±1.57)	74.60 (±2.53)
+ GconAttn	72.61 (±0.39)	68.59 (±0.96)	64.75 (±1.48)	71.80 (±1.21)
+ MHGRN	74.45 (±0.10)	71.11 (±0.81)	66.85 (±1.19)	80.60
+ QA-GNN	76.54 (±0.21)	73.41 (±0.92)	67.80 (±2.75)	82.77 (±1.56)
+ DGRN	78.20	74.00	69.60	84.10
+ GreaseLM	78.50 (±0.50)	74.20 (±0.40)	68.80 (±1.75)	84.80
+ JointLK	77.88 (±0.25)	74.43 (±0.83)	70.34 (±0.75)	84.92 (±1.07)
+ GSC	79.11 (±0.22)	74.48 (±0.41)	70.33 (±0.81)	86.67 (±0.46)
+ SAFE	76.93 (±0.37)	74.03 (±0.43)	69.20	87.13
+ HamQA	76.88	73.91	71.12	84.59
+ DRAGON^∗	-	76.00	72.00	-
+ DRAGON (w/o MLM)^∗	-	73.80	66.40	-
+ DHLK^∗	79.39 (±0.24)	74.68 (±0.26)	72.20 (±0.40)	86.00 (±0.79)
+ SEPTA (Ours)	79.61 (±0.17)	74.78 (±0.23)	72.33 (±0.35)	87.37 (±0.51)