Towards Usable Pattern Mining

About 15 years ago, pattern mining algorithms could extract millions of patterns from large amount of complex data in a few seconds. The goal of time efficiency had been achieved. However these millions of patterns were unusable because they were too numerous, redundant and of unequal interest. Since then, the new challenge is to make pattern mining methods “usable”.

This HDR presents my research activities on pattern mining, conducted over the last decade within the SemLIS team. I select two aspects of my work in this talk. The first aspect is about usable pattern mining for natural language processing tasks. The second aspect is about compression-based pattern mining approaches.