Paper Overview
Research area: Machine Learning Authors: Parviz Ghafariasl, Weimin Fu, Xiaolong Guo Published: 2026-09-03 arXiv: 2509.00004
Summary
Financial scams targeting older adults increasingly occur through text and voice channels such as email, SMS, and phone calls, unfolding over multiple conversational turns that begin with impersonation or casual contact, escalate through trust building and urgency, and culminate in requests for sensitive information or financial transfers. Because risk signals emerge incrementally across turns, effective detection requires models that continuously update risk estimates under resource-constrained deployment settings.
Key Contributions
- Cumulative turn-based risk assessment framework: incrementally aggregates conversational turns and re-estimates risk at each step, enabling dynamic scam monitoring across progressively evolving conversations.
- Multi-turn dialogue dataset: covers investment, charity, and tech-support scam scenarios; each conversation contains 2 to 8 turns, annotated at each cumulative stage with a qualitative risk level, a continuous risk score, explanatory rationales, and safety recommendations.
- Model evaluation: four small language models — Phi-4, LLaMA-3.2, DeepSeek-R1, and Qwen3 — were fine-tuned and evaluated under a unified training framework.
- Fine-tuned compact models capture fraud-related linguistic cues and cross-turn escalation patterns while retaining architectures suited to mobile and resource-constrained deployment.
- Among the evaluated models, Phi-4 and LLaMA-3.2 achieved the strongest turn-aware risk estimation performance relative to their parameter sizes.
- The results show that structured cumulative modeling can support deployment-oriented incremental scam risk assessment and highlight the potential of compact language models for privacy-aware, on-device fraud protection.