A Methodological View on Knowledge-Intensive Subgroup Discovery.pdfVIP

A Methodological View on Knowledge-Intensive Subgroup Discovery Martin Atzmueller and Frank Puppe University of Würzburg, 97074 Würzburg, Germany Department of Computer Science Phone: +49 931 888-6739, Fax: +49 931 888-6732 email: {atzmueller, puppe}@informatik.uni-wuerzburg.de Abstract. Background knowledge is a natural resource for knowledge-intensive methods: Its exploitation can often improve the quality of their results signifi- cantly. In this paper we present a methodological view on knowledge-intensive subgroup discovery: We introduce different classes and specific types of useful background knowledge, discuss their benefit and costs, and describe their appli- cation in the subgroup discovery setting. 1 Introduction Knowledge-intensive learning methods (e.g., [1]) use background knowledge for a sim- ple reason: Utilizing background knowledge can often significantly improve both the quality of their results and the efficiency of the search process. In this paper, we de- scribe how to exploit background knowledge for subgroup discovery, a method that has first been formalized by Kl?sgen [2] and Wrobel [3]: Subgroup discovery is a powerful and broadly applicable technique aiming at discovering interesting subgroups concern- ing a certain target property of interest, e.g., in the subgroup of smokers with a positive family history the risk of coronary heart disease (target property) is significantly higher than in the general population. Background knowledge can help to improve subgroup discovery in several ways, e.g., it can increase the representational expressiveness and also focus the subgroup discovery algorithm on the relevant patterns. Then, similar to a constrained query to a web search engine, the user is not flooded with too many (uninteresting) results. Fur- thermore, for increasing the efficiency of the search method the search space can often be constrained. However, knowledge acquisition is often challenging and costly, known as the ’knowledge acqui