Multilevel data summarization from information systems: a “rule + exception” approach

Issue title: Artificial Intelligence Advances in China

Article type: Research Article

Authors: Wang, Jue | Zhao, Min | Zhao, Kai | Han, Suqing

Affiliations: Institute of Automation, Chinese Academy of Sciences, Beijing 100080, China E‐mail: {Jue.wang, m.zhao, kai.zhao, suqing.han}@mail.ia.ac.cn

Abstract: In this paper we present a theoretical framework on Multilevel Data Summarization (MDS) – a process to summarize an information system into rule sets with different concise degrees (granularity) and corresponding exception sets, which is viewed as the rule‐plus‐exception model in cognitive science. In order to construct the theoretical framework of MDS, we propose the cognitive positive region and cognitive boundary region to substitute for the positive region and boundary region in rough set theory. Unlike current approaches, the structure of boundary region is paid more attention than positive region in this framework. Since exceptions are sometimes more important than rules for applications, we introduce a method to identify exceptions from a given information system, which concerns closely with the distribution of core attributes in the discernibility matrix.

Keywords: Data mining, rough set theory, granular computing, rule‐plus‐exception model

Journal: AI Communications, vol. 16, no. 1, pp. 17-39, 2003