Should Graph Convolution Trust Neighbors? A Simple Causal Inference Method

被引:32
|
作者
Feng, Fuli [1 ,2 ]
Huang, Weiran [3 ]
He, Xiangnan [4 ]
Xin, Xin [5 ]
Wang, Qifan [6 ]
Chua, Tat-Seng [2 ]
机构
[1] Sea NExT Joint Lab, Singapore, Singapore
[2] Natl Univ Singapore, Singapore, Singapore
[3] Chinese Univ Hong Kong, Hong Kong, Peoples R China
[4] Univ Sci & Technol China, Hefei, Peoples R China
[5] Univ Glasgow, Glasgow, Lanark, Scotland
[6] Google US, Mountain View, CA 94043 USA
基金
中国国家自然科学基金;
关键词
GCN; Causal Intervention; Model Inference;
D O I
10.1145/3404835.3462971
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Graph Convolutional Network (GCN) is an emerging technique for information retrieval (IR) applications. While GCN assumes the homophily property of a graph, real-world graphs are never perfect: the local structure of a node may contain discrepancy, e.g., the labels of a node's neighbors could vary. This pushes us to consider the discrepancy of local structure in GCN modeling. Existing work approaches this issue by introducing an additional module such as graph attention, which is expected to learn the contribution of each neighbor. However, such module may not work reliably as expected, especially when there lacks supervision signal, e.g., when the labeled data is small. Moreover, existing methods focus on modeling the nodes in the training data, and never consider the local structure discrepancy of testing nodes. This work focuses on the local structure discrepancy issue for testing nodes, which has received little scrutiny. From a novel perspective of causality, we investigate whether a GCN should trust the local structure of a testing node when predicting its label. To this end, we analyze the working mechanism of GCN with causal graph, estimating the causal effect of a node's local structure for the prediction. The idea is simple yet effective: given a trained GCN model, we first intervene the prediction by blocking the graph structure; we then compare the original prediction with the intervened prediction to assess the causal effect of the local structure on the prediction. Through this way, we can eliminate the impact of local structure discrepancy and make more accurate prediction. Extensive experiments on seven node classification datasets show that our method effectively enhances the inference stage of GCN.
引用
收藏
页码:1208 / 1218
页数:11
相关论文
共 50 条
  • [1] Causal Graph Inference
    Poilinca, Simona
    Parajuli, Jhanak
    Abreu, Giuseppe
    [J]. 2015 49TH ASILOMAR CONFERENCE ON SIGNALS, SYSTEMS AND COMPUTERS, 2015, : 1209 - 1213
  • [2] Entropic Causal Inference: Graph Identifiability
    Compton, Spencer
    Greenewald, Kristjan
    Katz, Dmitriy
    Kocaoglu, Murat
    [J]. INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 162, 2022,
  • [3] Causal Incremental Graph Convolution for Recommender System Retraining
    Ding, Sihao
    Feng, Fuli
    He, Xiangnan
    Liao, Yong
    Shi, Jun
    Zhang, Yongdong
    [J]. IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, 2024, 35 (04) : 4718 - 4728
  • [4] Causal Inference for Knowledge Graph Based Recommendation
    Wei, Yinwei
    Wang, Xiang
    Nie, Liqiang
    Li, Shaoyu
    Wang, Dingxian
    Chua, Tat-Seng
    [J]. IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, 2023, 35 (11) : 11153 - 11164
  • [5] Graph Convolution Networks based on Higher-order Trustable Neighbors
    He, Meixia
    Chen, Jianrui
    Li, Jiamin
    [J]. Proceedings - 2022 10th International Conference on Information Systems and Computing Technology, ISCTech 2022, 2022, : 215 - 221
  • [6] A2N: Attending to Neighbors for Knowledge Graph Inference
    Bansal, Trapit
    Juan, Da-Cheng
    Ravi, Sujith
    McCallum, Andrew
    [J]. 57TH ANNUAL MEETING OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS (ACL 2019), 2019, : 4387 - 4392
  • [7] Enhanced Graph Learning for Recommendation via Causal Inference
    Wang, Suhua
    Ji, Hongjie
    Yin, Minghao
    Wang, Yuling
    Lu, Mengzhu
    Sun, Hui
    [J]. MATHEMATICS, 2022, 10 (11)
  • [8] When Causal Inference Meets Graph Machine Learning
    Ma, Jing
    [J]. THIRTY-EIGHTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, VOL 38 NO 20, 2024, : 22676 - 22676
  • [9] Graph Construction using Principal Axis Trees for Simple Graph Convolution
    Alshammari, Mashaan
    Stavrakakis, John
    Ahmed, Adel F.
    Takatsuka, Masahiro
    [J]. arXiv, 2023,
  • [10] Causal Inference, is just Inference: A beautifully simple idea that not everyone accepts
    Rohde, David
    [J]. WORKSHOP AT NEURIPS 2021, VOL 163, 2021, 163 : 75 - 79