基于指令微调小语言模型的渐进式老年人金融诈骗增量风险评估
原标题：Incremental Risk Assessment of Progressive Elder Financial Scams via Instruction-Tuned Small Language Models
AI 摘要
Kansas State University的研究人员提出了一种基于累积对话轮次的老年人金融诈骗风险评估框架，通过逐步聚合对话内容并重新评估风险，实现实时监控。他们构建了包含投资、慈善和技术支持诈骗场景的多轮对话数据集，并微调了Phi-4、LLaMA-3.2、DeepSeek-R1和Qwen3等小型语言模型。实验表明，Phi-4和LLaMA-3.2在参数规模下表现出较强的风险估计性能，适合在移动和资源受限环境中部署，支持隐私保护的设备端欺诈防护。
正文节选
[Page 1] Incremental Risk Assessment of Progressive Elder Financial Scams via Instruction-Tuned Small Language Models Parviz Ghafariasl¹, Weimin Fu², Xiaolong Guo²*, Shing I. Chang¹* ¹Department of Industrial and Manufacturing Systems Engineering ²Department of Electrical and Computer Engineering Kansas State University, Manhattan, KS, USA Email: {parvizghafari, wei