PWAF : Path Weight Aggregation Feature for link prediction in dynamic networks

Abstract

The growing popularity of online social networks is obvious these days, and it provides researchers with an opportunity to find solutions for a variety of practical applications. The approach of recognizing network structure and finding missing and future links in a social network is known as link prediction. On time-varying or dynamic networks, the link prediction problem has two major challenges: accuracy and efficiency. The improvement of accuracy in link prediction problem in dynamic networks is the objective of this research. The goal of this paper is to propose the Path Weight Aggregation Feature (PWAF), which is a new feature based on ranking multi edge occurrences across the entire network. Different topological aspects of the networks (Local, Global, and Quasi−local) as well as Clustering Coefficient based features are taken into consideration for feature generation, in addition to the suggested Path Weight-Based Aggregation Feature (PWAF). One of the features used for better prediction is the Level−2 node clustering coefficient (CCLP2). Different machine learning models, such as Neural Network (NN), Logistic Regression (LR), XGBoost (XGB), Random Forest Classifier (RFC), and linear Discriminant Analysis (LDA), are evaluated and verified for link prediction. The experiments are carried out on seven different well-known dynamic networks data sets in terms of five performance evaluation metrics, including AUPR, F1-score, AVG PRECISION, BAL ACC SCORE, and AUC, and the results show that our proposed method and its variants outperform state-of-the-art methods. The results demonstrate that PWAF−RFC is the best performing variation out of all machine learning classifiers we have experimented with.