Evolution of Fuzzy Grammars to aid Instance Matching

Abstract

The need for information fusion exists in the semi-structured and unstructured domains - for example, to integrate responses from multiple sources into a unified response. This can be regarded as a two stage process - first to determine whether any two sources are considering the same real-world entities, and second, to ascertain how the attributes correspond (e.g. author/composer should correspond almost exactly to creator, business-location should correspond to address, etc). Within the unstructured and semi-structured attribute values there is frequently hidden structure -e.g. a free text attribute labeled as name might consist of title, first name and family name. Revealing this structure can greatly assist the matching process. In this paper, we outline a method for approximate matching of entities from different data sources and show how an evolutionary approach can create accurate approximate grammars to aid the information integration