Interesting Instance Discovery in Multi-Relational Data

Abstract

The general area of machine discovery focuses on methods to use computers to perform or assist discovery tasks. Herbert Simon described it as “gradual problemsolving processes of searching large problem spaces for incompletely defined goal objects” [Simon, 1995, p.171]. Today machine discovery research falls into two major categories, scientific discovery and knowledge discovery and data mining (KDD). In this paper we propose a new research direction that lies somewhere in-between these two trends: we call it interesting instance discovery (IID) which aims at discovering interesting instances in large, multi-relational datasets. There are three important characteristics for IID research: (1) Unlike scientific discovery and KDD, it aims at the discovery of particular interesting instances as opposed to general laws or patterns. (2) It is dealing with multi-relational data instead of numeric data that is best described as a relational graph or a semantic net. In such a network nodes represent objects and links represent relationships between them – see Figure 1 for an example of such a network from a bibliography domain. (3) Similar to KDD, it also focuses on data that are too large and complex to be analyzed manually by humans.