🤖 AI Summary
This study addresses the limitations of existing fraud detection systems, which typically analyze isolated messages and struggle to identify sophisticated scams that unfold over weeks or months through progressive trust-building dialogues. To tackle this challenge, the work proposes the first explainable agent architecture tailored for multi-category conversational fraud detection, integrating a summary-based memory mechanism with a dual-level (message- and conversation-level) detection framework that combines conversational semantic modeling and explainable AI techniques. The contributions include the release of ConScamBench-278, the first public multi-category benchmark dataset for conversational scams, achieving a 100% scam conversation detection rate on LoveFraud02 and 97.8% accuracy on ConScamBench-278. User studies further demonstrate that the system significantly enhances user trust and usability, attaining a usability score of 74.7.
📝 Abstract
Following the rapid progress of generative Artificial Intelligence, there is a growing threat posed by conversational scams. These scams often span over multiple weeks or months, gradually build trust and request for money or sensitive information. Existing scam-detection systems mainly focus on isolated messages, which renders them inadequate against this evolving threat. This paper extends single-message phishing detection and presents an explainable agentic system for detecting sophisticated conversational scams. It also introduces ConScamBench-278, an initial public multi-category benchmark for conversational scam detection spanning eight scam types, released to support reproducible evaluation and future expansion. On isolated messages the single-message detector attains 100% phishing recall, while the conversation-level detector identifies all conversational scams in the public LoveFraud02 corpus (83/83) and reaches 97.8% accuracy (95% CI [95.4, 99.0]) on ConScamBench-278. Two user studies (N = 100 and N = 45) further motivate the system: participants report frequently experiencing uncertainty when judging suspicious conversations. In an uncontrolled pre/post comparison, users self-reported trust, self-confidence, and perceived need for AI-based scam detection all increased (p < 0.001, Wilcoxon signed-rank). The system also receives a System Usability Scale score of 74.7 (95% CI [72.5, 76.9]), above the established usability benchmark.