Graph Negative Feedback Bias Correction Framework for Adaptive Heterophily Modeling

Jiaqi Lv; Qingfeng Du; Yu Zhang; Yongqi Han; Sheng Li

cs.LGMar 4, 2026

Graph Negative Feedback Bias Correction Framework for Adaptive Heterophily Modeling

Authors: Jiaqi Lv, Qingfeng Du, Yu Zhang, Yongqi Han, Sheng Li

Organizations: School of Computer Science and Technology, Tongji University, Shanghai, China

Abstract

Graph Neural Networks (GNNs) have emerged as a powerful framework for processing graph-structured data. However, conventional GNNs and their variants are inherently limited by the homophily assumption, leading to degradation in performance on heterophilic graphs. Although substantial efforts have been made to mitigate this issue, they remain constrained by the message-passing paradigm, which is inherently rooted in homophily. In this paper, a detailed analysis of how the underlying label autocorrelation of the homophily assumption introduces bias into GNNs is presented. We innovatively leverage a negative feedback mechanism to correct the bias and propose Graph Negative Feedback Bias Correction (GNFBC), a simple yet effective framework that is independent of any specific aggregation strategy. Specifically, we introduce a negative feedback loss that penalizes the sensitivity of predictions to label autocorrelation. Furthermore, we incorporate the output of graph-agnostic models as a feedback term, leveraging independent node feature information to counteract correlation-induced bias guided by Dirichlet energy. GNFBC can be seamlessly integrated into existing GNN architectures, improving overall performance with comparable computational and memory overhead.

Figures & tables

Figure 1: Schematic diagram of the negative feedback system and its adjustment process toward a new steady state. (a) denotes the training process, and (b) denotes the inference process.

Figure 2: GNFBC employs a graph-agnostic model to perform negative feedback correction during training. The graph-aware model and the graph-agnostic model share features and weights. During the inference stage, GNFBC follows the standard inference process.

	Homophilic Graphs					Heterophilic Graphs
	Cora	CiteSeer	PubMed	Computers	Photo	Chameleon	Squirrel	Texas	Cornell
RF	79.53	68.28	84.56	73.42	77.25	53.92	38.76	85.07	82.48
MLP	71.39	66.58	82.29	70.46	78.61	46.27	31.55	81.63	83.92
SGC	69.74	58.34	81.56	75.28	81.88	64.57	43.17	55.26	46.83
GCN	75.18	67.89	85.43	81.89	89.32	60.78	45.33	55.03	61.82
GAT	76.89	67.56	83.46	82.13	90.32	63.46	43.11	58.98	60.03

Table 1: Performance comparison on five homophilic and four heterophilic graph datasets. The best result is shown in bold , while the second best is marked with underline .

	YelpChi		Amazon
	AUC	F1	AUC	F1
RF	86.49	70.24	94.33	91.50
MLP	69.25	54.86	92.76	90.02
SGC	61.27	48.58	80.13	76.59
GCN	61.03	48.32	80.49	76.93
GAT	60.39	47.53	83.87	79.20

Table 2: Performance comparison on YelpChi and Amazon datasets in terms of AUC and F1-macro (%).

Figure 3: Performance comparison between GNFBC and standalone Graph-aware or Graph-agnostic model on homophilic and heterophilic datasets.

	GNFBC	Graph-aware	Graph-agnostic
YelpChi	91.34	80.87	69.25
Amazon	97.55	94.01	92.76

Table 3: Performance comparison between GNFBC and standalone Graph-aware or Graph-agnostic model on YelpChi and Amazon datasets.

Figure 4: Performance of SGC, GCN, and GAT models with the GNFBC framework on the YelpChi dataset

Figure 5: Performance of SGC, GCN, and GAT models with the GNFBC framework on the Texas dataset

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Dataset	Nodes	Edges	Classes
Cora	2,708	5,429	7
CiteSeer	3,327	4,732	6
PubMed	19,717	44,338	3
Computers	13,752	245,861	10
Photo	7,650	119,081	8
Wisconsin	251	499	5

Appendix

Table 4: Summary of Dataset Statistics

Explore similar work

Layer Embedding Deep Fusion Graph Neural Network
Apr 25, 2026Taihua Xu, Genhao Tian, Jicong Fan +3Graph Structure Learning Graph Neural Networks
Jiangsu University of Science and Technology · The Chinese University of Hong Kong, Shenzhen · Chongqing University of Post and Telecommunications
Inductive Subgraphs as Shortcuts: Causal Disentanglement for Heterophilic Graph Learning
Apr 21, 2026Xiangmeng Wang, Qian Li, Haiyang Xia +3Disentangled Representation Learning Graph Neural Networks
Curtin University Perth, Australia · The Education University of Hong Kong New Territories, Hong Kong
Revealing the Pitfalls and Re-Evaluating the Advancement of Heterophilic Graph Learning
Sep 9, 2024Sitao Luan, Qincheng Lu, Chenqing Hua +3Graph Neural Networks Node Classification
Mila - Quebec Artiﬁcial Intelligence Institute · 1McGill University

Apr 25, 2026J/K move · Enter open · S save

Layer Embedding Deep Fusion Graph Neural Network

Taihua Xu, Genhao Tian, Jicong Fan, Xibei Yang+2

Jiangsu University of Science and Technology · The Chinese University of Hong Kong, Shenzhen · Chongqing University of Post and Telecommunications

Graph Structure Learning Graph Neural Networks Graph Representation Learning Multi-Scale Feature Fusion Graph Heterophily

Open paper AskPDF arXiv

Graph Neural Networks (GNNs) have demonstrated impressive performance in learning representations from graph-structured data. However, their message-passing mechanism inherently relies on the assumption of label consistency among connected nodes, limiting their applicability to low-homophily settings. Moreover, since message passing operates as a hierarchical diffusion process, GNNs face challenges in capturing long-range dependencies. As network depth increases, the structural noise along heterophilic edges tends to be amplified, resulting in over-smoothing. This issue becomes especially prominent in highly heterophilic graphs, where the propagation of inconsistent semantics across the topology continually exacerbates misaggregation. To address this issue, we propose a novel framework named Layer Embedding Deep Fusion Graph Neural Network (LEDF-GNN). Specifically, we design a Layer Embedding Deep Fusion (LEDF) operator that nonlinearly fuses multi-layer embeddings to capture inter-layer dependencies and effectively alleviate deep propagation degradation. Meanwhile, to mitigate structural heterophily, LEDF-GNN employs a Dual-Topology Parallel Strategy (DTPS) that simultaneously leverages the original and reconstructed topologies, allowing for adaptive structure-semantics co-optimization under diverse homophily conditions. Extensive semi-supervised classification experiments on the citation and image benchmarks demonstrate that, under both homophilic and heterophilic settings, LEDF-GNN consistently outperforms state-of-the-art baselines, validating its effectiveness and generalization capability across diverse graph types.

Abstract

Explore similar work

Layer Embedding Deep Fusion Graph Neural Network

Inductive Subgraphs as Shortcuts: Causal Disentanglement for Heterophilic Graph Learning

Revealing the Pitfalls and Re-Evaluating the Advancement of Heterophilic Graph Learning