On Segmentation of Interaction Values - Developing Concepts in Applied Intelligence - page 27

On Segmentation of Interaction Values

Nguyen Chi Lam 1 , Hiep Xuan Huynh 1 , and Fabrice Guillet 2

1 College of Information and Communication Technology Cantho University, Vietnam
2 Polytechnic School of Nantes University, France

Abstract. Post-processing of association rules with interestingness
measures is considered as one of the most dicult and interesting task in
the research domain of Knowledge Discovery from Databases (KDD). In
this paper, we propose a new approach to discover the behaviors of in-
terestingness measures by modeling the interaction between them. The
interaction values are calculated based on the capacity function (also
called fuzzy measure) and then are segmented to discover the interac-
tion's trends of clusters of interestingness measures.

Keywords: Knowledge Discovery from Databases, association rules, in-
terestingness measures, correlation graph, capacity function, Sugeno
measure, interaction value, segmentation.

1 Introduction

The analysis of the relations between interestingness measures based on the cal-
culations of interestingness values from a data set structured by rules, especially
association rules [1], is a challenge in Knowledge Discovery in Databases (KDD).
Each interestingness measure is often considered as representing an aspect or a
property on the data set studied. The process of determining the approriated
interestingness measures for a data set always a complex problem to resolve. In
addition, the use of only one interestingness measure to reflect the whole data
set is very dicult. In reality, each data set has itself many aspects that are not
express obviously.
In this paper, we deal with a new approach based on the interaction between
interestingness measures[3] [4] [7] , using the a graph model called interaction
graph. After the determination of clusters of interestingness measures and the
selection of the approriate interestingness measures that are strongtly represent
the specific aspects of the data set, we can easily rank the association rules
generated from the data set according to the order of the common interestingness
measure obtained from the group of interestingness measures chosen.

2 Interestingness Measures

Suppose that we have a data set D contains the transactions in a supermarket
[1]. The set of items (itemset) I = corresponds with a set of avail-

able articles in a supermarket. The set X, contains the articles appeared in a
transaction. In fact, the transaction expresses the bought articles of custormers.

{i 1 ,i 2 , ..., i n }