A new constraint for mining sets in sequences

International audienceDiscovering interesting patterns in event sequences is a popular taskin the field of data mining. Most existing methods try to do thisbased on some measure of cohesion to determine an occurrence of apattern, and a frequency threshold to determine if the pattern occursoften enough. We introduce a new constraint based on a newinterestingness measure combining the cohesion and the frequency of apattern. For a dataset consisting of a single sequence, the cohesion is measured as the average length of thesmallest intervals containing the pattern for each occurrence of itsevents, and the frequency is measured as the probability of observingan event of that pattern. We present a similarconstraint for datasets consisting of multiple sequences. We presentalgorithms to efficiently identify the thus defined interestingpatterns, given a dataset and a user-defined threshold. After applyingour method to both synthetic and real-life data, we conclude that itindeed gives intuitive results in a number of applications