A Node Classification Method Based on Structural Correlation and Semantic Similarity in Networks

doi:10.1007/s13369-026-11573-8

Return

A Node Classification Method Based on Structural Correlation and Semantic Similarity in Networks

2026-08-20

PRE

Sajjad Ahmad

Shamoona Jabeen *

Syed Shahid Shah

DOI：10.1007/s13369-026-11573-8

Original

Original request for help

Save

Abstract

En 中文

In the era of rapidly expanding networked data, extracting meaningful node representations from sparsely labeled graphs remains a critical challenge. Graph learning methods are widely used for semi-supervised node classification. This paper proposes a scalable hybrid approach that combines structural correlation and semantic similarity to improve node classification under limited labeled data. The method incorporates label propagation to leverage unlabeled nodes in dense regions and introduces a mutual information-guided pseudo-labeling strategy to prioritize informative nodes. Experiments on six real-world datasets (BAT, EAT, UAT, PubMed, Cora, and CiteSeer) demonstrate consistent superiority over strong baselines, with the proposed hybrid method achieving up to 91.6% classification accuracy and outperforming the strongest competing approach across all benchmark datasets. We also provide a theoretical justification for combining the two similarity sources and demonstrate the convergence of the proposed propagation scheme. These findings highlight the effectiveness of the proposed approach for reliable knowledge extraction and decision-making in complex networked systems with scarce annotations.

Keywords:

Semi-supervised learning

Similarity

Correlation

Label propagation