Search

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. [ABSTRACT FROM AUTHOR]