Mining quantitative association under inequality constraints

Abstract

The problem of discovering association rules was first introduced in 1994 by R. Agrawal and R. Stikant. In the past several years, there has been much active work in developing algorithms for mining association rules. However, in discovering the patterns, it has been realized that not all associations are of interest. It is more desirable if a user can limit the target associations by specifying different constraints. For example, a marketing personnel may only want to know which items are often sold together with a total price more than 200. That is, he is interested in association rules which satisfy a given inequality constraint for a set of quantitative items. The aim of our work is to research for new methods and algorithms to extract subtle and embedded knowledge in the database satisfying inequality constraints. Three types of constraints are considered in our work for different data item relationships. The first type of constraints are the inequality constraints which consider the quantitative relationships between items. The second type of constraints are the temporal constraints which consider the temporal and quantitative relationships between items. The last type of constraints are the taxonomy constraints which consider the multi-layer relationships between items. In our work, we consider arithmetic inequality constraints which are composed of common operators such as (+, - ,* ,/). We believe they are the most common constraints and can be easily extended to other queries such as the max( ), min( ), and avg( ). Finding the interesting associations is not the only objective of our work, we also attempt to simplify and speed up the whole mining process by making use of the arithmetic properties of the input constraints. Finally, preliminary experimental results of the proposed algorithms are also reported and discussed.