Interactive mining and evaluation of pattern sets.

We live in the era of data and need tools to reveal valuable information in large amounts of data. The goal of exploratory data mining is to provide as much insight in given data as possible. Within this field, pattern set mining aims at revealing structure in the form of pattern sets. Individual patterns describe subsets of the data; a non-redundant set of patterns provides a global description.

Although pattern set mining is a promising line of research, it is not yet widely adopted. The goal of this project is to make pattern set mining practically useful by enabling the user to explore the data and identify interesting structure, where interestingness is specified interactively.

The task is formalised as learning problem, such that machine learning techniques can be used to learn a user-specific interestingness distribution on pattern sets. In each iteration the system presents pattern sets to the user, who ranks these. Starting from a prior based on the Minimum Description Length principle, the interestingness distribution is updated using this feedback. To allow for smooth interaction, sampling is used to efficiently obtain candidate solutions.

Finally I propose to address the lack of methods for qualitative evaluation of pattern sets. By exploiting user feedback and measures from information retrieval, evaluation comes at little extra cost. Expected result of this project is a comprehensive system for the interactive exploration of discrete data, revealing structure of interest to the domain expert.