Modeling and Mining the Rule Evolution
Temporal data mining attempts to provide accurate information about an evolving business domain. A framework is proposed to discover continuously temporal knowledge based on a session model. The main concepts and properties in temporal rule induction are defined and proved in a formal way, using first-order linear temporal logic. The measures of first-order rule are used to discover evolutional regularity about the rule. The mining process consists of four stages: planning, session mining, merge mining, and post-processing. Various session mining for temporal data generates a measure sequence of first-order rule. The parameter estimation method applicable to the measure sequence with a small-sample is presented, based on the principle of information diffusion. Experiment shows the validity and simplicity of the method.
KeywordsInformation Diffusion Session Mining Diffusion Estimation Measure Sequence Rule Evolution
Unable to display preview. Download preview PDF.
- 1.Agrawal, R., Psaila, G.: Active data mining. In: Proc. of the KDD 1995, pp. 3–8. AAAI Press, California (1995)Google Scholar
- 5.Spiliopoulou, M., Roddick, J.F.: Higher Order Mining. In: Proc. 2nd Intl. Conf. on Data Mining Methods and Databases, Cambridge, UK, pp. 309–320 (2000)Google Scholar
- 7.Pan, D., Shen, J.: Ontology Service-based Architecture for Continuous Knowledge Discovery. In: Proc. of the 4th ICMLC 2005, pp. 2155–2160. IEEE Press, Guangzhou (2005)Google Scholar