FUSE: Fast Semi-Supervised Node Embedding Learning via Structural and Label-Aware Optimization
Sujan Chakraborty, Rahul Bordoloi, Anindya Sengupta, Olaf Wolkenhauer, Saptarshi Bej
TL;DR
FUSE addresses node classification in graphs without feature vectors by learning embeddings from structural and label signals. It unifies three objectives—a linear modularity-based unsupervised term, supervised intra-class compactness, and semi-supervised label propagation with attention—into a single gradient-based optimization with QR orthonormalization. Empirical results show FUSE achieves accuracy on par with or better than state-of-the-art methods while significantly reducing computational cost, including scalability to large graphs like ArXiv. The framework is particularly effective when embeddings must be generated quickly or updated incrementally in dynamic or resource-constrained environments. Overall, FUSE provides a feature-free, fast, and scalable pathway to high-quality node embeddings for semi-supervised classification tasks.
Abstract
Graph-based learning is a cornerstone for analyzing structured data, with node classification as a central task. However, in many real-world graphs, nodes lack informative feature vectors, leaving only neighborhood connectivity and class labels as available signals. In such cases, effective classification hinges on learning node embeddings that capture structural roles and topological context. We introduce a fast semi-supervised embedding framework that jointly optimizes three complementary objectives: (i) unsupervised structure preservation via scalable modularity approximation, (ii) supervised regularization to minimize intra-class variance among labeled nodes, and (iii) semi-supervised propagation that refines unlabeled nodes through random-walk-based label spreading with attention-weighted similarity. These components are unified into a single iterative optimization scheme, yielding high-quality node embeddings. On standard benchmarks, our method consistently achieves classification accuracy at par with or superior to state-of-the-art approaches, while requiring significantly less computational cost.