Hybrid Sensing Approach for Real-Time Chat Behavior Detection and Classify Social Engineered Attacks

View at publisher Summarise with VIRAT
Authors: Yasas Manujaya Ekanayake, L. Ranathunga
Cited by: 0
References: 14

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

Social engineering attacks in real-time chat environments exploit psychological manipulation to extract sensitive information, posing significant risks to individuals and organizations. Traditional detection methods often struggle with high latency, limited contextual awareness, and poor interpretability, rendering them inadequate for dynamic conversational settings. This paper proposes a novel hybrid system integrating lightweight rule-based pattern matching with machine learning, leveraging DistilBERT embeddings and engineered features for accurate, low-latency detection of social engineering attacks. A rolling fraud score mechanism captures multi-turn conversational threats, while a non-recursive architecture ensures memory efficiency. The proposed method has been evaluated using a composite of real and semi-synthetic multi-turn chat datasets, the hybrid model achieved consistent message and conversation-level performance with $\tilde 93\% $ accuracy and low latency (

Fields of Study & Subjects

Spam and Phishing Detection Adversarial Robustness in Machine Learning Hate Speech and Cyberbullying Detection

Publication details

Journal / source
IEEE International Conference on Industrial and Information Systems (ICIIS)
Pages
19–24
Publication date
16 January 2026
Language
EN