Towards Human-like Perception: Learning Structural Causal Model in Heterogeneous Graph

Abstract

Heterogeneous graph neural networks have become popular in various domains. However, their generalizability and interpretability are limited due to the discrepancy between their inherent inference flows and human reasoning logic or underlying causal relationships for the learning problem. This study introduces a novel solution, HG-SCM (Heterogeneous Graph as Structural Causal Model). It can mimic the human perception and decision process through two key steps: constructing intelligible variables based on semantics derived from the graph schema and automatically learning task-level causal relationships among these variables by incorporating advanced causal discovery techniques. We compared HG-SCM to seven state-of-the-art baseline models on three real-world datasets, under three distinct and ubiquitous out-of-distribution settings. HG-SCM achieved the highest average performance rank with minimal standard deviation, substantiating its effectiveness and superiority in terms of both predictive power and generalizability. Additionally, the visualization and analysis of the auto-learned causal diagrams for the three tasks aligned well with domain knowledge and human cognition, demonstrating prominent interpretability. HG-SCM's human-like nature and its enhanced generalizability and interpretability make it a promising solution for special scenarios where transparency and trustworthiness are paramount.

arXiv comment28 pages, 10 figures, 6 tables, accepted by Information Processing & Management

References in corpus (27)

• Decoupled Weight Decay Regularization
• Shortcut Learning in Deep Neural Networks
• Why social networks are different from other types of networks
• Graph Contrastive Learning with Augmentations
• Deep Graph Library: A Graph-Centric, Highly-Performant Package for Graph Neural Networks
• GNNExplainer: Generating Explanations for Graph Neural Networks
• Graph Transformer Networks
• Discovering Symbolic Models from Deep Learning with Inductive Biases
• Higher-Order Explanations of Graph Neural Networks via Relevant Walks
• DAGs with NO TEARS: Continuous Optimization for Structure Learning
• Parameterized Explainer for Graph Neural Network
• Wikidata as a seed for Web Extraction
• DAG-GNN: DAG Structure Learning with Graph Neural Networks
• Causal Discovery with Reinforcement Learning
• A comparison of two approaches for measuring interdisciplinary research output: the disciplinary diversity of authors vs the disciplinary diversity of the reference list
• Heterogeneous Graph Representation Learning with Relation Awareness
• Deep Structural Causal Models for Tractable Counterfactual Inference
• Explainability Techniques for Graph Convolutional Networks
• Reinforced Causal Explainer for Graph Neural Networks
• CausPref: Causal Preference Learning for Out-of-Distribution Recommendation
• Causality-based CTR Prediction using Graph Neural Networks
• DAGs with No Fears: A Closer Look at Continuous Optimization for Learning Bayesian Networks
• Interpretable and Generalizable Graph Learning via Stochastic Attention Mechanism
• CASTLE: Regularization via Auxiliary Causal Graph Discovery
• Nonlinear Causal Discovery with Confounders
• A Local Method for Identifying Causal Relations under Markov Equivalence
• Differentiable DAG Sampling