Identity and Reference in web-based Knowledge Representation (IR-KR)

Accurate information extraction program is a key prerequisite for the correct identification of entities on the Web, but their development is not at all a trivial task. Moreover, the maintenance, optimization, and customization of such programs require significant effort and resources. Recent trends in the Information Extraction (IE) research introduce the vision of algebraic information extraction. An important aspect of this vision is the declarative description of the extraction flows, outside of the monolithic IE program, by using a small set of generic operators. In our approach, we follow this vision and address the three main requirements that have never been addressed together in the same system before. First, we introduce a new methodology for efficient entity extraction from unstructured data, which involves the mapping of the extracted entities to already existing structured or semi-structured data. Second, we propose a set of operators addressing a comprehensive set of IE tasks, such as extracting atomic elements and aggregating them to complex real world objects, identifying relationships. Third we propose operators for leveraging global identifier providers, such as OKKAM. To verify our approach, we have implemented an information extraction system and evaluated our operators on a real example extraction flow for retrieving product information from forum pages.