Pattern Discovery in Temporal Databases: Some Recent Results

In their influential paper [MTV95], Mannila et al. formulated the problem of discovering frequently occurring temporal patterns in sequences, where temporal patterns are specified using the notion of an episode [MTV95]. It has been recognized by Padmanabhan and Tuzhilin [PT96] that temporal patterns can be specified using temporal logic and that this method generalizes the episodes approach proposed in [MTV95] in two ways. First, temporal logic is more expressive than the episodes formalism. Therefore, it allows to specify a richer set of temporal patterns. Second, the approach proposed in [PT96] is applicable not only to sequences (i.e., propositional case), but also to temporal databases (the first-order case). It is also described [PT96] how frequent occurrences of patterns expressed with temporal logic can be discovered using temporal logic programming methods [AM89]. In particular, [PT96] uses an observation that any temporal logic formula can be simulated with a temporal logic program. Therefore, in order to find frequently occurring temporal patterns belonging to a certain class of temporal logic formulas, one has to generate an appropriate temporal logic program that simulates these formulas and counts the number of occurrences of these patterns.