Explainable and Robust Detection of AI Bot-Generated Social Dialogues Using a Hybrid Feature-Attention Framework

View at publisher Summarise with VIRAT
Authors: Janith Hathnagoda, L. Ranathunga
Cited by: 0
References: 7

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This research paper aims to address significant challenges posed by rapidly evolving conversational AI bots in distinguishing them from genuine humans in social dialogs. Traditional detection methods relying on static patterns or simple heuristics fall short against advanced adaptive bots. A novel hybrid approach is proposed, which combines deep explainability and adversarial robustness to effectively detect conversational bots. The method employs meticulous data preprocessing, extracting numeric features and semantic BERT embeddings processed through a dual-branch neural network enhanced with attention mechanisms. Adversarial training through embedding perturbations strengthens robustness against subtle textual manipulations, while interpretability is ensured through attention-based insights and SHAP analysis, which led it to create transparent decision rules. The experimental results demonstrate approximately ninety-six percent accuracy, significantly outperforming existing methods. This hybrid framework uniquely balances explainability, robustness, scalability, real-time capabilities, and advanced bot detection in dynamic conversational environments.

Fields of Study & Subjects

Spam and Phishing Detection Adversarial Robustness in Machine Learning Authorship Attribution and Profiling

Publication details

Journal / source
IEEE International Conference on Industrial and Information Systems (ICIIS)
Pages
108–113
Publication date
16 January 2026
Language
EN