Last changed: August 4, 2004
FactMine: fact and Ontology Mining for Question Answering
The goal of the proposed project is to develop unsupervised methods for the extraction of ontological information from texts. We will examine three techniques: natural language processing, pattern extraction, and clustering. We expect to employ suitable versions of these techniques for expanding the Dutch version of EuroWordNet and thus create an ontology which can be used in the open domain question answering part of the IMIX Demonstrator. We will also apply these techniques to texts of a restricted domain in order to evaluate their usefulness for such domains.
This proposal fits in the paradigm of the Semantic Web, an effort for adding semantic annotation to online texts. For storing facts and relations the project will use the Resource Description Framework (RDF) and the Web Ontology Language (OWL), both developed within the Semantic Web effort.
By providing techniques for generating ontologies and other knowledge sources, the present proposal relates to three of the thematic priorities which were defined to be relevant for the IMIX Demonstrator: clarification dialogues, which require ontologies for detecting problems in the questions, architectural problems caused by slow response time as a result of large collection document search, for which we propose an alternative, and answer construction from multiple documents.
- Maarten Marx
- Maarten de Rijke
- Erik Tjong Kim Sang
- November 1, 2004