User Identity Linkage on Social Networks: A Review of Modern Techniques and Applications

TL;DR

This paper reviews recent methods for linking user identities across social networks, discussing techniques, challenges, and applications to aid researchers in understanding and improving cross-platform user identification.

Contribution

It provides a comprehensive overview of recent UIL techniques, including problem formulations, feature extraction, algorithms, datasets, and evaluation metrics from 2016 onward.

Findings

Summarizes main problem formulations in UIL

Details feature extraction strategies and algorithms

Highlights datasets and evaluation metrics used

Abstract

In an Online Social Network (OSN), users can create a unique public persona by crafting a user identity that may encompass profile details, content, and network-related information. As a result, a relevant task of interest is related to the ability to link identities across different OSNs. Linking users across social networks can have multiple implications in several contexts both at the individual level and at the group level. At the individual level, the main interest in linking the same identity across social networks is to enable a better knowledge of each user. At the group level, linking user identities through different OSNs helps in predicting user behaviors, network dynamics, information diffusion, and migration phenomena across social media. The process of tying together user accounts on different OSNs is challenging and has attracted more and more research attention in the last fifteen years. The purpose of this work is to provide a comprehensive review of recent studies (from 2016 to the present) on User Identity Linkage (UIL) methods across online social networks. This review aims to offer guidance for other researchers in the field by outlining the main problem formulations, the different feature extraction strategies, algorithms, machine learning models, datasets, and evaluation metrics proposed by researchers working in this area. The proposed overview takes a pragmatic perspective to highlight the concrete possibilities for accomplishing this task depending on the type of available data.