Random Walks with Variable Restarts for Negative-Example-Informed Label Propagation

TL;DR

This paper introduces CusTaRd, a novel label propagation algorithm that uses variable restarts and negative sampling to better incorporate negatively labeled nodes, improving classification accuracy on graph datasets.

Contribution

The paper proposes a new reformulation of random walks with restarts, enabling variable restart probabilities and negative sampling, which enhances label propagation performance.

Findings

CusTaRd outperforms existing algorithms in benchmark tests.

Variable restarts improve the ability to avoid negatively-labeled nodes.

Negative sampling from neighbors enhances predictive accuracy.

Abstract

Label propagation is frequently encountered in machine learning and data mining applications on graphs, either as a standalone problem or as part of node classification. Many label propagation algorithms utilize random walks (or network propagation), which provide limited ability to take into account negatively-labeled nodes (i.e., nodes that are known to be not associated with the label of interest). Specialized algorithms to incorporate negatively labeled samples generally focus on learning or readjusting the edge weights to drive walks away from negatively-labeled nodes and toward positively-labeled nodes. This approach has several disadvantages, as it increases the number of parameters to be learned, and does not necessarily drive the walk away from regions of the network that are rich in negatively-labeled nodes. We reformulate random walk with restarts and network propagation to enable "variable restarts", that is the increased likelihood of restarting at a positively-labeled node when a negatively-labeled node is encountered. Based on this reformulation, we develop CusTaRd, an algorithm that effectively combines variable restart probabilities and edge re-weighting to avoid negatively-labeled nodes. In addition to allowing variable restarts, CusTaRd samples negatively-labeled nodes from neighbors of positively-labeled nodes to better characterize the difference between positively and negatively labeled nodes. To assess the performance of CusTaRd, we perform comprehensive experiments on four network datasets commonly used in benchmarking label propagation and node classification algorithms. Our results show that CusTaRd consistently outperforms competing algorithms that learn/readjust edge weights, and sampling of negatives from the close neighborhood of positives further improves predictive accuracy.