偏序约简框架下增量自学习推理:税务稽查的例子

Incremental Self-learning Reasoning in Briefed Partially-ordered Case Base: A Tax Inspection Case

本文讨论欺诈防范领域中税务稽查的例子。在相关文献基础上分析了目前线性推理的不足,提出构造偏序约简范例集,给出了CBR循环过程中范例获取、记忆、扩容、推理等算法,由此实现范例推理机增量自学习机制。算法相比线性检索和记忆有着较高的性能和准确度,在税务稽核选案、信用卡欺诈、公司财务数据审计方面都可以有相当广阔的应用。

Tax inspection is a special category in fraud detection area. Based on the investigation on related works, this paper discusses the weaknesses of current linear reasoning in case-based reasoning context. We propose a partially-ordered briefed case base, under which algorithms of case retrieval, retain, adaptation in CBR circle are offered. This enables an incremental self-learning mechanism on target cases. Experimental result has shown advantages over linear reasoning mechanism. This reasoning machine can be widely embeded in tax inspection, credit card fraud detection, and financial auditing applications.