📄 中文摘要
该研究提出了一种神经符号方法用于欺诈检测，重点关注如何在推理阶段监测概念漂移而无需标签。模型通过符号规则编码了对欺诈的知识，例如 V14 低于某个阈值意味着欺诈。然而，当这种关系开始变化时，规则能否作为预警信号？研究探讨了神经符号概念漂移监测的有效性，强调了在没有标签的情况下，如何利用符号规则来捕捉潜在的变化。文章还提到相关的背景架构，以帮助理解这一机制。
神经符号欺诈检测：在 F1 降低之前捕捉概念漂移（无标签）
📄 English Summary
Neuro-Symbolic Fraud Detection: Catching Concept Drift Before F1 Drops (Label-Free)
This study presents a neuro-symbolic approach for fraud detection, focusing on how to monitor concept drift during inference without labels. The model encodes its knowledge of fraud through symbolic rules, such as V14 below a certain threshold indicating fraud. However, what happens when this relationship begins to change? Can the rules serve as a warning signal? The research explores the effectiveness of neuro-symbolic concept drift monitoring, emphasizing how symbolic rules can be utilized to capture potential changes in the absence of labels. The article also references related architectural backgrounds to aid in understanding this mechanism.