A New ILP-based Concept Discovery Method for Business Intelligence

In this work, we propose a multi-relational concept discovery method for business intelligence applications. Multi-relational data mining finds interesting patterns that span over multiple tables. The obtained patterns reveal useful information for decision making in business environments. However, as the patterns include multiple relations, the search space gets intractably complex. In order to cope with this problem, various search strategies, heuristics and language pattern limitations are employed in multi-relational learning systems. In this work, we develop an ILP-based concept discovery method that uses inverse resolution for generalization of concept instances in the presence of background knowledge and refines these patterns into concept definitions by applying specialization operator. There are two main benefits in this approach. The first one is to relax the strong declarative biases and user-defined specifications. The second one is to integrate the method on relational databases so that usage of the system is facilitated in business intelligence applications.