An inductive database for mining temporal patterns in event sequences
Alexandre Vautier, Marie-Odile Cordier, René Quiniou · 2005
Abstract. An inductive database formalizes data mining as looking for interesting patterns by querying a database containing raw data and patterns. We propose an inductive database extension for mining temporal patterns in event sequences where few types of event are present and so temporal information is of major importance for information extraction. We are interested in temporal patterns, named chronicles, which are sets of events such that the delay between their occurrences is bounded by a numerical interval. We propose a generality relation defined on this kind of pattern and related adaptations to the version space algorithm used for learning. The method can be used to extract temporal patterns that discriminate long event sequences where times of events represent most of the useful information. This extraction process is achieved by answering queries that set (maximum and minimum) frequency thresholds on the temporal patterns in events sequences. 1