Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adversarial Botometer: Adversarial Analysis for Social Bot Detection

Shaghayegh Najari, Davood Rafiee, Mostafa Salehi, Reza Farahbakhsh

TL;DR

This work tackles the challenge of detecting social bots under adversarial, human-like manipulation by GenAI. It introduces Adversarial Botometer, a GAN-inspired framework where a Seq2Seq bot generator competes with a Contextual-LSTM detector in adversarial training, including a pre-training phase and an MCST-guided minimax objective. The evaluation on Cresci 2017 and Midterm 2018 spans three scenarios—adversarial play, data poisoning, and domain/model explanations—revealing that detector robustness varies across bot types and domains and highlighting the need for dynamic, adversarially trained defenses. The study further uses SHAP to interpret feature importance and compares multiple modeling approaches, offering practical guidance for building more resilient bot-detection systems in real-world, evolving OSN environments.

Abstract

Social bots play a significant role in many online social networks (OSN) as they imitate human behavior. This fact raises difficult questions about their capabilities and potential risks. Given the recent advances in Generative AI (GenAI), social bots are capable of producing highly realistic and complex content that mimics human creativity. As the malicious social bots emerge to deceive people with their unrealistic content, identifying them and distinguishing the content they produce has become an actual challenge for numerous social platforms. Several approaches to this problem have already been proposed in the literature, but the proposed solutions have not been widely evaluated. To address this issue, we evaluate the behavior of a text-based bot detector in a competitive environment where some scenarios are proposed: \textit{First}, the tug-of-war between a bot and a bot detector is examined. It is interesting to analyze which party is more likely to prevail and which circumstances influence these expectations. In this regard, we model the problem as a synthetic adversarial game in which a conversational bot and a bot detector are engaged in strategic online interactions. \textit{Second}, the bot detection model is evaluated under attack examples generated by a social bot; to this end, we poison the dataset with attack examples and evaluate the model performance under this condition. \textit{Finally}, to investigate the impact of the dataset, a cross-domain analysis is performed. Through our comprehensive evaluation of different categories of social bots using two benchmark datasets, we were able to demonstrate some achivement that could be utilized in future works.

Adversarial Botometer: Adversarial Analysis for Social Bot Detection

TL;DR

This work tackles the challenge of detecting social bots under adversarial, human-like manipulation by GenAI. It introduces Adversarial Botometer, a GAN-inspired framework where a Seq2Seq bot generator competes with a Contextual-LSTM detector in adversarial training, including a pre-training phase and an MCST-guided minimax objective. The evaluation on Cresci 2017 and Midterm 2018 spans three scenarios—adversarial play, data poisoning, and domain/model explanations—revealing that detector robustness varies across bot types and domains and highlighting the need for dynamic, adversarially trained defenses. The study further uses SHAP to interpret feature importance and compares multiple modeling approaches, offering practical guidance for building more resilient bot-detection systems in real-world, evolving OSN environments.

Abstract

Social bots play a significant role in many online social networks (OSN) as they imitate human behavior. This fact raises difficult questions about their capabilities and potential risks. Given the recent advances in Generative AI (GenAI), social bots are capable of producing highly realistic and complex content that mimics human creativity. As the malicious social bots emerge to deceive people with their unrealistic content, identifying them and distinguishing the content they produce has become an actual challenge for numerous social platforms. Several approaches to this problem have already been proposed in the literature, but the proposed solutions have not been widely evaluated. To address this issue, we evaluate the behavior of a text-based bot detector in a competitive environment where some scenarios are proposed: \textit{First}, the tug-of-war between a bot and a bot detector is examined. It is interesting to analyze which party is more likely to prevail and which circumstances influence these expectations. In this regard, we model the problem as a synthetic adversarial game in which a conversational bot and a bot detector are engaged in strategic online interactions. \textit{Second}, the bot detection model is evaluated under attack examples generated by a social bot; to this end, we poison the dataset with attack examples and evaluate the model performance under this condition. \textit{Finally}, to investigate the impact of the dataset, a cross-domain analysis is performed. Through our comprehensive evaluation of different categories of social bots using two benchmark datasets, we were able to demonstrate some achivement that could be utilized in future works.
Paper Structure (20 sections, 4 equations, 9 figures, 2 tables, 1 algorithm)

This paper contains 20 sections, 4 equations, 9 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Bot Generation
  4. Bot Detection
  5. Existing Research GAP and Objective
  6. Proposed Method: Adversarial Botometer
  7. Scenario 1: Adversarial Game between bot and bot bot detection
  8. Pre-training Phase
  9. Adversarial Training
  10. Scenario2: Data Poisoning
  11. Scenario3: Domain and Model Explanation
  12. Experimental Results
  13. Datasets
  14. scenario 1
  15. scenario 2
  16. ...and 5 more sections

Figures (9)

  • Figure 1: Proposed Adversarial Game
  • Figure 2: Training procedure of bot in the context of GAN (Perplexity vs. Cosine similarity of generated samples to the real ones
  • Figure 3: Training procedure of bot detection in the context of GAN.
  • Figure 4: Training speed of bot and bot detection in the context of GAN
  • Figure 5: Probability distribution of three classes of social bots for attack examples selected from data.
  • ...and 4 more figures