Domain-driven in-depth pattern discovery: A practical methodology
Longbing Cao, R Schurmann, Chengqi Zhang · UTS ePRESS (University of Technology Sydney) · 2005
Traditional data mining is a data-driven trial-and-error process. The patterns discovered via predefined models in the above process are generic patterns. Generally, they are often not really interesting to constraint-based real business, hi order to work out patterns that are of interest and actionable to the real world, in-depth patterns are often essential. This type of pattern discovery is more likely to be a business or industry domain-driven human-machine-cooperated process. The use of in-depth patterns requires the development of a more practical methodology, than is presently available for guiding real-world data mining. This paper proposes such a practical data mining methodology, referred to as domain-driven in-depth pattern discovery (DDID-PD). The main idea of the DDID-PD methodology is to mine in-depth patterns through domain-driven iterative human-machine interaction in a constraint-based context. Using this methodology as a basis, we demonstrate some of our work in mining in-depth correlations in Australian Stock Exchange (ASX) data and preliminary research on developing a quality knowledge base for Centrelink interventions. The deployment of DDID-PD to ASX data mining tasks has shown that the methodology is practical and has potential for further improving the analysis of large quantities of data to identify patterns for practical use by industry and business. © 2013.