A novel DFS/BFS approach towards link prediction

TL;DR

This paper introduces a new link prediction method combining centrality measures with machine learning, addressing biases in traditional approaches and showing promising results especially with degree centrality and random node selection.

Contribution

It presents a novel approach that integrates centrality measures with classical machine learning for link prediction, expanding beyond existing bias-prone methods.

Findings

Effective when using degree centrality

Performs well with randomly selected nodes

Shows potential for further research

Abstract

Knowledge graphs have been shown to play a significant role in current knowledge mining fields, including life sciences, bioinformatics, computational social sciences, and social network analysis. The problem of link prediction bears many applications and has been extensively studied. However, most methods are restricted to dimension reduction, probabilistic model, or similarity-based approaches and are inherently biased. In this paper, we provide a definition of graph prediction for link prediction and outline related work to support our novel approach, which integrates centrality measures with classical machine learning methods. We examine our experimental results in detail and identify areas for potential further research. Our method shows promise, particularly when utilizing randomly selected nodes and degree centrality.