Mini Proposal

Topic:

• Semi-auto ontology building for a domain specific ontology.

Approach:

1. Filter out datasets to include only triples for the domain of interest.
2. Identify the relation type (part-whole or generalization), by:
   1. categorizing (classifying) the part-of-speech of the predicate (property)
3. Use result from 2 as input to pyowl module for building OWL classes (ontology).

Datasets:

• Knowledge Bases from NELL, DBpedia, and Linked Data in a triple format.

Tools:

• Data cleaning, filtering, categorizing, and classifying:

  • NLP techniques (py: genism, nltk, patterns.en)

• Building OWL classes and axioms:

  • my pyowl (OWL mapping/transformation module)

Related:

• Semi-Automated Ontology building through Natural Language Processing, http://cirworld.org/journals/index.php/ijct/article/view/908N
• http://www.heikopaulheim.com/docs/ijswis_2014.pdf NELL, dbpedia, and owl
• Generating Ontologies from Linked Data GOLD

Ref:

• Identify relations types and semantics with Deep Learning, see slide: 21
• Towards Semi-automatic Ontology Building Supported by Large-scale Kno wledge Acquisition pdf
• An introduction to ontology learning, by Jens Lehmann and Johanna Voelker
  source: http://aksw.org/Groups/MOLE.html