Use of algorithms for a user specific reduction of amounts of interesting association rules

Abstract

The huge amounts of stored digital data, which are nowadays available in many domains contain lots of previously unknown coherences. For the user it is often difficult or unfeasible to find those coherences he is interested in without any technical support. One kind of these coherences are association rules. This paper presents an iterative procedure which supports the user in finding these association rules he is interested in, by considering his interests without the explicit formulation of these interests by the user in advance. The procedure presents iteratively association rules to the user, who has to value each of them as interesting or uninteresting. With the help of a genetic algorithm the procedure learns interactively the interests of the user and formulates classification rules, which are used to classify the not yet presented association rules in the classes interesting and uninteresting so that only interesting classified association rules are presented to the user in the following. The procedure was evaluated on a standard dataset and a dataset of the web2.0 application flick. The evaluation results show, that the developed procedure is useful for both standard database applications and innovative web2.0 applications. Different genetic methods and scenarios of interests were evaluated. The most interesting evaluation results will be presented in this paper.