Detecting Twitter Bots with Machine Learning and Propagation Graph Analysis

Publication Date

1-1-2023

Document Type

Conference Proceeding

Publication Title

Proceedings - 2023 5th International Conference on Transdisciplinary AI, TransAI 2023

DOI

10.1109/TransAI60598.2023.00022

First Page

87

Last Page

90

Abstract

Twitter is a social networking platform that offers microblogging services to millions of accounts every day. However, not all of these accounts are operated by actual (real) individuals. Many accounts are controlled by automated software, commonly referred to as 'bots'. While Twitter doesn't explicitly prohibit the presence of bots, it does establish rules that govern their usage. Twitter encourages the use of bots for positive purposes, such as sharing helpful information and enhancing the user experience. On the other hand, malicious activities like spamming or harassing users are strictly prohibited. The identification of Twitter bots has become increasingly crucial, particularly in preventing the spread of misinformation and maintaining online discourse quality. Existing methods for detecting Twitter bots primarily rely on feature-based and text-based techniques. In our research, we propose a graph-based approach to identify Twitter bots and as a first step employ classical network analysis to examine the behavior of bot accounts within the Twittersphere. The paper outlines the process of data collection and graph propagation construction and conducts network analysis on the real and bot accounts propagation graphs. Additionally, a machine-learning model utilizing the Weisfeiler-Lehman graph kernel for graph classification is presented. The findings obtained from classical network analysis reveal significant differences in network structures between real user relationship graphs and bot relationship graphs. The results from various machine learning models in this paper also indicate that the graph-based model is more effective at distinguishing bot relationship graphs from real user graphs.

Funding Number

22-RSG-08-034

Keywords

graph classification, graph kernels, machine learning, Network Analysis, Twitter Bot

Department

Computer Science

Link to Full Text

Share

COinS