Centrality-guided random walks for balanced node representation in graph embeddings
Graph-based representations provide an effective means for modeling complex real-world problems characterized by long-tailed degree distributions. A small fraction of nodes exhibit dense connectivity, while the majority remain only sparsely connected. Nodes with limited connections often suffer from degraded embedding quality. The imbalance is particularly salient in random walk-based graph embedding techniques because the sampling process systematically favors densely linked neighborhoods. To address the sampling bias induced by degree disparity, we introduce Centrality-Guided Random Walks, which adaptively sample nodes based on centrality-driven transition probabilities. Our model integrates multiple centrality metrics to balance the exploration of both highly connected and sparsely connected nodes. The adjustment produces robust embeddings for all nodes, regardless of their degree. Experimental results demonstrate that the proposed method improves overall classification accuracy with notable gains among tail-node groups. They underscore the effectiveness of incorporating node centrality into walk-based graph embedding for long-tail scenarios.
Authors
- Mucheol Kim (ORCID: https://orcid.org/0000-0003-2682-2195)
- Yejong Ryu
- Mincheol Shin (ORCID: https://orcid.org/0000-0001-8983-1553)
- Ka Lok Man (ORCID: https://orcid.org/0000-0002-5787-4716)
- Tae-Young Choe (ORCID: https://orcid.org/0000-0002-5253-3488)
Institutions
- Chung-Ang University (KR)
- Xi’an Jiaotong-Liverpool University (CN)
Publication Details
- Journal
- Journal Of Big Data
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1186/s40537-026-01603-5
- Primary Topic
- Advanced Graph Neural Networks
- Type
- article
- Field-Weighted Citation Impact
- 0.00