Empirical Analysis of Unsupervised Link Prediction Algorithms in Weighted Networks

Abstract

Complex relationships in many real-world problems can be represented as networks where nodes represent individuals and relationships among them are represented by links. So, one of the key issues in such networks is the evolution or creation of edges. Link prediction is the solution where it finds the missing links (edges) in a static case (in a given snapshot of a network) or future links in a dynamic case (given several snapshots of networks at different time instants). In this experimental work, we consider the weighted versions of different existing algorithms and showed their performance on several real networks of different domains. We observed that the weighted version of the method path of length 3, i.e. L3-WT outperforms other methods on all four evaluation metrics with some exceptions. LHN1-WT method is the second outperformer on LesMiserables and Netscience datasets. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.