PQCLP: Parameterized quantum circuit based link prediction in dynamic networks

Abstract

Link prediction has been challenging, especially when the network is dynamic and complex. The most effective classical method for performing this task involved using machine learning algorithms with features taken from topological network indices. Even while these traditional ML algorithms perform better, they still require a lot of processing resources as the size and number of features in the network increase. This is the ideal situation where quantum computing may fit, as it provides impressive predictions and speedup arising out of quantum phenomena like superposition, entanglement, parallelization, and high dimensional space. Additionally, relatively little research has been done to examine the full potential of quantum computation for link prediction. A few of the earlier attempts are limited to projecting the features to quantum space and then using quantum-projected kernels with classical ML techniques or using hybrid classifiers by incorporating quantum enhancement in traditional random walks. We propose Parameterized Quantum Circuit based Link Prediction (PQCLP) model where we have used quantum circuits not only for projecting the classical data but also for training and optimization in quantum space using Variational Circuits Aka Ansatz which is a parameterized circuit. Here we employ two quantum methods namely Variational Quantum Classifier (VQC) and Quantum Neural Network Classifier (QNN) having classical equivalence with Support Vector Machines and Neural Networks respectively. We present here a detailed comparison of these models with their classical counterparts, within different feature categories and test ratios, and finally with a few state-of-the-art methods using several performance evaluation metrics. © 2024 Elsevier B.V.