Dynamic Fuzzy Sampler for Graph Neural Networks

Graph Neural Networks (GNNs) are widely used across fields, with inductive learning replacing transductive learning as the mainstream training paradigm due to its superior memory efficiency, computation speed, and generalization. Neighbor node sampling, a key step in inductive learning, is critical to model performance. However, existing samplers focus only on adjacency matrix-based sampling, neglecting the varying impacts of different neighbors on target nodes over time. They usually aggregate the neighbor information in a simple way such as averaging or summing, which limits the information representation, robustness, and generalization. To address these limitations, we propose a Dynamic Fuzzy Sampler (DFS) based on a Gaussian fuzzy system. DFS accounts for node diversity and models the uncertainties and fuzziness in neighbor-target mutual information dynamically. Specifically, DFS first innovatively constructs a learnable Gaussian fuzzy set system for determining the membership degree of different neighbors to the target node at different moments. Subsequently, DFS aggregates the target node embeddings and membership-weighted neighbor embeddings to update the target node's features, which makes the target node utilize the sampling information more effectively. The aggregated target node effectively captures the graph structure information and neighbor node information, which can facilitate the subsequent graph neural network-based graph representation model with stronger representation and generalization capabilities. Experimental results on supervised and self-supervised graph datasets demonstrate that DFS consistently outperforms state-of-the-art sampling schemes. DFS achieves up to 1.90% and 9.52% F1-score improvement compared to the state-of-the-art schemes on small- and large-scale graphs, respectively.