关联规则分析中兴趣度量Lift与Conviction的关系探讨及教育数据验证  

Exploration of the Relationship between Interest Measures Lift and Conviction in Association Rule Analysis and Education Data Validation

作　　者：万鑫 李梓如 李裕梅[1] 

机构地区：[1]北京工商大学数学与统计学院,北京

出　　处：《数据挖掘》2024年第3期189-206,共18页Hans Journal of Data Mining

摘　　要：关联规则分析是数据挖掘中最常用的研究方法之一。在关联关系的发现过程中兴趣度量是关联规则发现的理论基础,它可以度量规则的重要程度,其中Lift和Conviction这两个度量在数据分析中被广泛应用于筛选关联规则。本文对这两种兴趣度量进行了研究。首先,提出并证明了当后项集固定时,Conviction取值随Lift取值单调增加,且Conviction (Lift)是一个凸函数。然后,证明了当Confidence固定时,Conviction取值随Lift取值单调增加,且Conviction (Lift)是一个凹函数。最后,综合以上两个方面,得到一个重要结论:当后项集保持不变或当Confidence固定时,根据Conviction和Lift筛选出来的规则都是相同的。最后,利用某高校数学类专业三个年级的成绩数据进行了定理及相应结论的验证。Association rule analysis is one of the most active research methods in data mining. In the process of finding association relationships, interest measures are the theoretical basis and can measure the significance of rules, where Lift and Conviction are widely used in data analyses to find association rules. This paper studies these two measures. First, it is proven that when the Consequent is fixed, the value of Conviction increases monotonically with the value of Lift, and Conviction is a convex function of Lift. Second, when Confidence is fixed, the value of Conviction increases monotonically with the value of Lift, and Conviction is a concave function of Lift. Then, integrating the above two aspects, we obtain an important conclusion: when the Consequent remains fixed or when the Confidence value is fixed, the rules selected by Conviction are the same as those selected by Lift. Finally, the theorems and the corresponding conclusion are verified by using the achievement data of three grades of mathematics major in a university.

关 键 词：关联规则分析 LIFT CONVICTION 函数关系 数据分析验证 

分 类 号：G63[文化科学—教育学]