Using Network Analysis to Identify Genuine and Fraudulent Social Media Profiles
Last night I received a follow request on Instagram from Sharon_lee881116. It didn’t take me long to determine the account was fake, or at least was not close enough to me or my network for me to accept their request. As our digital social lives become a larger part of everyday life, we have to learn how to navigate a vast network of genuine and malicious users. Over time, certain characteristics of social media accounts have begun to form strong indicators for making this determination. For example, when an Instagram user navigates to another account, the application will show a list of users who follow the account navigated to and are followed by the initial user. When I am determining whether an account is real, I look at follower ratio. In general, the people I am following are following me back, and the same is generally true for them, too. Therefore, most of the accounts I engage with have follower-following ratios closer to 1. In theie article, “Identifying Fake Accounts on Social Networks Based on Graph Analysis and Classification Algorithms,” Mohammadreza Mohammadrezaei, Mohammad Ebrahim Shiri, and Amir Masoud Rahmani combine and further graph research aimed at detecting ingenuine users.
The researchers used a labeled dataset to create and test an algorithm that, when the dataset was balanced to include an artificial proportion of fake accounts (relative to real ones), performed with high accuracy. They use friendship graphs, which are, for every node, the graph of nodes to which they are connected. Additionally, they used many features in their algorithm, including common friends, defined as the total number of shared users between two accounts and Jaccard similarity, which is a coefficient representing the similarity of two users, or nodes, friendship networks. Mohammadrezaei et al were able to address problems of previous research by resampling their data, as well as by using PCA and machine learning. These allowed the researchers to achieve extremely high accuracy rates.
The researchers were able to formalize the same type of process individual users go through when they receive follow requests. While the researchers were attempting to create a classification algorithm that could detect fake accounts through network analysis using a large dataset, further research could provide users with scientifically proven methods that reduce the risk one is allowing a malicious account to follow them and can be done from a given social media application by an individual in the moment, rather than through expensive computation.
https://www.hindawi.com/journals/scn/2018/5923156/#conclusion-and-future-work