Charles VERNEREY Thesis defense

The objective of this thesis is to develop innovative approaches to effectively represent user preferences in multi-objective decision contexts for the extraction of interesting knowledge.
Firstly, we introduce a new Constraint Programming (CP) model to efficiently extract Pareto optimal patterns (also known as skypatterns) that scales effectively.
Next, we demonstrate how skypatterns can be leveraged to extract high-quality, non-redundant association rules without the need to set thresholds, in contrast to existing state-of-the-art approaches.
We then go beyond Pareto optimality to represent user preferences by introducing a novel approach based on the Choquet integral, a complex aggregation function that accounts for interactions (positive or negative) between the measures used to evaluate pattern quality. Lastly, we present a new Java-based CP library for modeling and solving several pattern mining problems.