Using SAT and SQL for Pattern Mining in Relational Databases

Abstract. In this paper, we present an ongoing work bridg-ing the gap between pattern mining, SQL and SAT for a particular class of patterns. We extend the work presented in [2] that proposes a logical query language for rule patterns satisfying Armstrong’s axioms. Our contributions are the fol-lowing: firstly, we allow a large part of the relational tuple calculus (SQL) to be used in the specification of queries. Sec-ondly, we propose a boolean encoding of the query that can be used to compute answers even in the case of non Armstrong-compliant queries. Some experiments have been performed on top of Derby (embedded Java DBMS) and a modified version of MiniSat to show the feasibility of the approach.