Authors: Mohammad Majid Akhtar, Navid Shadman Bhuiyan, Rahat Masood, Muhammad Ikram, Salil S. Kanhere

Published on: February 06, 2024

Impact Score: 8.2

Arxiv code: Arxiv:2402.03740

Summary

• What is new: A novel framework, BotSSCL, is proposed to enhance the detection of sophisticated social bots in online social networks by leveraging self-supervised contrastive learning.
• Why this is important: Existing social bot detection models struggle with sophisticated bots that mimic real users and show poor generalizability across different datasets.
• What the research proposes: The BotSSCL framework employs contrastive learning to improve the separation between bots and humans in embedding space, enhancing detection robustness and generalizability.
• Results: BotSSCL outperforms existing methods with approx. 6% and 8% higher F1 scores on two datasets, demonstrates 67% F1 in cross-dataset tests, and reduces successful adversarial evasion to 4%.

Technical Details

Technological frameworks used: Self-Supervised Contrastive Learning

Models used: BotSSCL

Data used: nan

Potential Impact

Online social networks and platforms could greatly benefit from this technology by improving the integrity of their user interactions and protecting against information manipulation.

Want to implement this idea in a business?

We have generated a startup concept here: BotGuard.