基于网络的数值关联规则挖掘方法

关联规则 ( Association Rules)发现的是属性间的关系 .属性可以是逻辑型的 ,也可以是数值型的 .在从逻辑型属性描述的数据中发现关联规则方面已经有许多比较成熟的算法 ,而在数值型属性方面则不然 .将数值关联规则挖掘问题映射成逻辑关联规则挖掘问题是一种方便有效的方法 .本文给出了一个新的数值属性关联规则挖掘算法 ,该算法利用数据本身的特征决定对数值属性值域的划分 ,进而将划分后的所有区间映射为逻辑属性 (项目 ) ,在此基础上可以挖掘出更容易理解、更具有概括性的有效关联规则 .本文给出了一个发现频繁项目集搜索算法 ,并采用一种纵向数据库格式来简化项目集支持度的计算.

A Lattice-based Mining Algorithm for Quantitative Association Rules

The association rules discovery the relations among the attributes. An attribute can be Boolean or quantitative. There are lots of algorithms for mining Boolean association rules, but few for quantitative. It is an efficient and convenient method to mapping quantitative attributes into Boolean attributes. A new algorithm for mining quantitative association rules is presented in this paper. Quantitative attribute values are partitioned into basic intervals according to the their distribution in the database, and if possible, the adjacent basic itervals will be merged. Then the intervals are mapped into Boolean attributes (i.e., the items) in this way. More understandable, general interesting quantitative association rules can be mined. The algorithm uses a new searching process to enumerate frequent itemsets. Furthermore, the algorithm uses a vertical database format to compute the support of each itemset, where each value or interval is associated with a list of records in which it occurs.