Publications & Resources
Unsupervised Ontology Generation From Unstructured Text
Hamid Mousavi, Deirdre Kerr and Markus R. Iseli
Ontologies are a vital component of most knowledge acquisition systems, and recently there has been a huge demand for generating ontologies automatically since manual or supervised techniques are not scalable. In this paper, we introduce OntoMiner, a rule-based, iterative method to extract and populate ontologies from unstructured or free text. OntoMiner transforms text into a graph structure called a textGraph in which nodes are candidate terms and words from the text and edges are grammatical, semantic, and categorical relations between nodes. OntoMiner iteratively uses graph pattern rules over the textGraphs to mine ontological information and at the end of each iteration, based on the newly found information, OntoMiner improves the existing ontology. Our preliminary experiments indicate that OntoMiner achieves up to 93.4% accuracy, which to our knowledge exceeds the accuracy levels of previous work.
Mousavi, H., Kerr, D., & Iseli, M. R. (2013). Unsupervised ontology generation from unstructured text (CRESST Report 827). Los Angeles: University of California, Los Angeles, National Center for Research on Evaluation, Standards, and Student Testing (CRESST).