Multi-Relational Data Mining Based on Higher-Order Inductive Logic Programming

This paper presents a novel multi-relational data mining (MRDM) approach from a perspective of considering higher-order inductive logic programming to dealing with the representation formalism problems of existing multi-relational data mining approaches. In our approach, examples, background knowledge,hypotheses and target concepts are represented in Escher, a higher-order logic programming language.Escher can describe semantically complicated data and patterns, and explicitly supports a variety of data types, including graph. Moreover, our approach explores and exploits the techniques of HILP based on Escher to efficiently construct search space and proposal a novel methodology of MRDM.Furthermore, we present an architecture for efficiency and scalability of MRDM based on HILP. We believe that our approach based on higher-order inductive logic programming will has a key role to play in the growth of MRDM while several major call for algorithms that explicitly exploit the semantically complicated and topological substructures of data.