Feature based graph augmentation improves learning on incomplete graphs
Learning the Latent Structure: A Feature-Centric Approach to Graph Data Augmentation
Machine Learning
Summary
Graphs are helpful for showing how things connect, but real-world graphs often miss some connections, which makes it hard for computers to learn from them. The authors created a new method that works with the features of nodes rather than guessing missing connections directly. This method helps computers figure out hidden links by learning from the patterns in node data, even when labels or supervision are limited. Tested on many different graphs, their method consistently performs better and runs faster than previous approaches.
What this means in practice
- •For data science teams: Improve machine learning on incomplete or sparse graph datasets by generating more informative node features without relying on existing labels.
- •For cybersecurity analysts: Enhance detection models by better handling missing or incomplete network connection data through refined node representations.
Authors
Yu Song, Zhigang Hua, Yan Xie, Bingheng Li, Jingzhe Liu, Bo Long, Jiliang Tang, Hui Liu
Abstract
Graph-structured data plays a pivotal role in modeling complex relationships. However, real-world graphs are often incomplete due to data collection and observational constraints, severely limiting the effectiveness of modern graph learning pipelines. While existing Graph Data Augmentation (GDA) methods attempt to refine graph structures for improved downstream performance, they are typically label-dependent, computationally expensive, and inherently transductive, limiting their applicability in practical scenarios. In this work, we present a novel feature-centric graph data augmentation framework that bypasses explicit structure modeling by operating directly in the embedding space. Through a self-supervised inverse masking process, our method captures latent ties between observed and complete graphs, enabling recovery of unobserved structural signals through refined node representations. To enhance robustness under noisy and sparse supervision, we introduce a message regularizer and a bootstrap strategy for effective training and generalization. Evaluated on ten graph datasets spanning multiple domains, our approach, SelfAug, consistently outperforms state-of-the-art methods in both accuracy and efficiency across inductive and cold-start settings, highlighting its potential as a scalable and generalizable solution for real-world graph learning scenarios.