Improving personalized link prediction by hybrid diffusion

TL;DR

This paper introduces a hybrid heat conduction and random walk algorithm for personalized link prediction in social networks, improving diversity and accuracy over classical methods.

Contribution

It generalizes heat conduction to personalized link prediction, incorporates a ground node for performance enhancement, and proposes hybrid algorithms outperforming existing methods.

Findings

Outperforms classical similarity-based algorithms in diversity and accuracy.

Adding a ground node significantly improves heat conduction performance.

Hybrid algorithms outperform others in AUC, precision, recall, and Hamming distance.

Abstract

Inspired by traditional link prediction and to solve the problem of recommending friends in social networks, we introduce the personalized link prediction in this paper, in which each individual will get equal number of diversiform predictions. While the performances of many classical algorithms are not satisfactory under this framework, thus new algorithms are in urgent need. Motivated by previous researches in other fields, we generalize heat conduction process to the framework of personalized link prediction and find that this method outperforms many classical similarity-based algorithms, especially in the performance of diversity. In addition, we demonstrate that adding one ground node who is supposed to connect all the nodes in the system will greatly benefit the performance of heat conduction. Finally, better hybrid algorithms composed of local random walk and heat conduction have been proposed. Numerical results show that the hybrid algorithms can outperform other algorithms simultaneously in all four adopted metrics: AUC, precision, recall and hamming distance. In a word, this work may shed some light on the in-depth understanding of the effect of physical processes in personalized link prediction.