Med Onto Medical Ontology Learning System Work in

Med. Onto: Medical Ontology Learning System (Work in Progress) Syed Farrukh Mehdi Reza Fathzadeh S. M. Faisal Abbas (Presenter) {fmehdi, reza, fabbas}@cs. dal. ca 1

Introduction: � Ontology ◦ Machine readable information � Text ◦ Human readable information, most of the current information is text. � Ontology Learning ◦ (Semi) automatic extraction of relevant concept and relations � Medical Domain 2

Methodology � Syntax based concept learning augmented with domain specific subject corpora Syntax Based Extraction Domain Specific Knowledge base 3

Domain Specific Corpus � Medical Domain Terminology ◦ Open. Galen project �GALEN Terminology Server � For Other domains, domain specific terminology corpus should be used. 4

Syntax Based Extraction Levels Paul Buitelaar 5

Term Extraction � Parsing ◦ Linguistic Method �Using Production Rules specified by linguists ◦ Statistical Method �Using statistical models derived from written text. � We used Stanford NLP Parser which is a statistical parser � Dependency Trees instead of Parse Trees 6

Synonym Extraction � Domain Specific Terminology Corpus � Language corpus for general concepts ◦ GRAIL Terminology Server for Medical Domain ◦ Word. Net for English Language 7

Concept Extraction � Intension ◦ Formal and information definition of terms � Extension ◦ Deriving concepts � Linguistic Realization ◦ Concept coverage 8

Terminal and Compound Concepts � Terminal Concept ◦ Nouns, Noun Phrases � Compound Concepts ◦ Defined Rules 9

Relation Extraction � Concepts are related � Defined Rules 10

Rules (IN) � IN subordinating conjunction (FUNC_WORD) or preposition (PREP) ◦ “of” � Candidate for Taxonomy 11

Rules (CC) � CC coordinating conjunction ◦ “and”, “or” etc ◦ Compound concepts, broken into terminal concepts 12

Rules (RB, DT, PDT) � RB adverb and adverbial phrase � DT determiner/demonstrative pronoun � Ignored in our work so far 13

Rule (VB) � Verb is used as a relation between subject and object 14

Rule (JJ+NN -> NP) � JJ adjective � NN common noun 15

Algorithm � Recursive, until dependency tree is exhausted � Create compound concepts and relate them with the rule and then apply the rules on the sub phrases 16

Other Work Framework Institution Reference ASIUM INRIA, Jouy--‐en--‐Josas Faure and Nedellec 1999 Text. To. Onto AIFB, University of Karlsruhe Madche and Volz 2001 HASTI Amir Kabir University, Teheran Shamsfard, Barforoush 2004 Onto. LT DFKI, Saarbrucken Buitelaar et al. 2004 DOODLE Shizuoka University Morita et al. 2004 Text 2 Onto AIFB, University of Karlsruhe Cimiano and Volker 2005 Onto. Learn University of Rome Velardi et al. 2005 OLE Brno University of Technology Novacek and Smrz 2005 Onto. Gen Institute Jozef Stefan, Ljubljana Fortuna et al. , 2007 GALe. On Technical University of Madrid Manzano-Macho et al. 2008 DINO DERI, Galway Novacek et al. 2008 Onto. Lancs Lancester University Gacitua et al. 2008 RELEx. O AIFB, University of Karlsruhe Volker and Rudolph 2008 Onto. Comp University of Dresden Sertkaya 2008 17

References � [Buitelaar 05] Paul Buitelaar, etal. Ontology Learning from Text, October 3 rd , 2005 � [Kim 09] Jin-Dong Kim et al. , Overview of Bio. NLP’ 09 Shared Task On Event Extraction � [Stuck] Semantic Technologies, Ontology Learning, Prof. Dr. Heiner Stuckenschmidt, Dr. Johanna Völker � [Biemann] Chris Biemann: Ontology Learning from Text: A Survey of Methods � [Stan. Parser] http: //nlp. stanford. edu/software/lexparser. shtml � [Word. Net] http: //wordnet. princeton. edu/ � [Open. GALEN] http: //www. opengalen. org/ 18

Thank you. Please provide us Comments and Directions 19