Structure Discovery from Sequential Data

In this paper we describe I-Subdue, an extension to the Subdue graph-based data mining system. I-Subdue operates over sequentially received relational data to incrementally discover the most representative substructures. The ability to incrementally refine discoveries from serially acquired data is important for many applications, particularly as computer systems become more integrated into human lives as interactive assistants. This paper describes initial work to overcome the challenge of locally optimal substructures overshadowing those that are globally optimal. We conclude by providing an overview of additional challenges for sequential structure discovery.

• Type: Conference
• Year: 2004
• Where: FLAIRS
• Authors: Jeffrey Coble, Diane J. Cook, Lawrence B. Holder, Runu Rathi
• Keywords: Artificial Intelligence, FLAIRS 2004, Optimal Substructures, Representative Substructures, Subdue Graph-based Data