✨ TL;DR
F²LP-AP is a training-free label propagation method that adapts to both homophilous and heterophilous graphs using geometric median prototypes and local clustering coefficients. It achieves competitive accuracy with GNNs while being significantly more computationally efficient.
Semi-supervised node classification on graphs is a fundamental task, but state-of-the-art Graph Neural Networks suffer from two major limitations: they require expensive iterative training with multi-layer message passing, and they rely on strong homophily assumptions that fail on heterophilous graphs where dissimilar nodes are connected. Existing training-free methods like traditional Label Propagation lack the adaptability to handle diverse graph structures effectively.
F²LP-AP constructs robust class prototypes using the geometric median rather than simple averaging, making the method more resilient to outliers. It dynamically adjusts propagation parameters based on the Local Clustering Coefficient (LCC) to adapt to local graph topology, enabling the framework to handle both homophilous and heterophilous structures. The method operates without gradient-based training, making it computationally efficient.
What the paper shows.
F²LP-AP achieves competitive or superior accuracy compared to trained GNNs across diverse benchmark datasets while significantly outperforming existing baselines in computational efficiency. The method demonstrates effectiveness on both homophilous and heterophilous graph structures without requiring gradient-based training.
The paper does not explicitly detail specific accuracy numbers or computational time comparisons in the abstract. The scalability to very large graphs and the sensitivity of the method to the Local Clustering Coefficient threshold are not discussed. The choice of geometric median and its computational cost relative to simpler alternatives is not thoroughly analyzed.
✨ Generated by Claude · Apr 25, 2026 · Read the PDF for authoritative content.