对最小置信度门限的置疑  

Doubts about Min Confidence Threshold

作　　者：马占欣[1,2] 王新社[1] 黄维通[3] 陆玉昌[4] 

机构地区：[1]周口师范学院计算机科学系 [2]清华大学智能系统与技术国家重点实验室北京100084 [3]清华大学计算机科学与技术系 [4]清华大学智能系统与技术国家重点实验室

出　　处：《计算机科学》2007年第6期216-218,共3页Computer Science

基　　金：国家自然科学基金(60473115)。

摘　　要：在关联规则挖掘中,通常使用最小支持度和最小置信度两个门限来衡量一条规则是不是一个强规则。本文对最小置信度这个参数的实际意义,从理论和实践上进行了分析研究和探讨,发现使用最小置信度进行限制不仅所挖掘出的规则质量较低,还有可能遗漏一些具有重要价值的规则,进一步提出提升率比置信度更能反映实际情况,在关联规则挖掘中改用最小支持度和最小提升率作为衡量准则,其结论更加准确,意义也更明确。The two thresholds： min support and min confidence are often used to evaluate if a rule is a strong rule or not in association rules mining. The present paper analyzes and explores the practical significance of min confidence both theoretically and practically. It finds that not only the quality of the mined rules is comparatively low but also some important rules are probably missed out if rain confidence is used to restrict. Different from previous research, this paper proposes that upgrade rate can reflect the actuality more accurately than rain confidence and the result will be more accurate and clearer if min support and min upgrade are taken as the weighing rule in association rules mining.

关 键 词：数据挖掘 关联规则 兴趣度 置信度 提升率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]