Relational databases form the backbone of modern information systems, yet significant business logic remains hidden within the data itself—implicit regularities that schemas fail to capture. This thesis introduces a deterministic framework to mine these patterns.
We present MATILDA and MAHILDA, two systems capable of discovering expressive Tuple-Generating Dependencies and Horn Rules directly from relational tables. Prioritizing Explainable AI, we generate verifiable, rule-shaped artifacts that support auditing, data quality, and logical imputation without the opacity of black-box models.
A scalable algorithm to discover first-order dependencies with multi-atom heads and existential witnesses, capturing complex implications across multiple tables.
A specialized system for mining recursive Horn rules. It introduces disjoint semantics to strictly prevent vacuous self-loops, serving as a rigorous baseline for ILP.