Learning of Semantic Relations between Statistical Techniques - PowerPoint PPT Presentation

Learning of Semantic Relations between Ontology Concepts using Learning of Semantic Relations between Statistical Techniques Ontology Concepts using Statistical A. Tegos Techniques Introduction The Proposed Method Finding the Semantic A. Tegos 1 , 2 , V. Karkaletsis 1 , A. Potamianos 2 Relations of concepts Finding the Cardinality tegos@iit.demokritos.gr, vangelis@iit.demokritos.gr, Restrictions Experimental potam@telecom.tuc.gr Assessment Conclusions 1 Institute of Informatics and Telecommunications, NCSR “Demokritos”, Future Plans Greece 2 Department of Electronics and Computer Engineering, Technical University of Crete, Greece High-level Information Extraction Workshop 2008 (HLIE08), ECML-PKDD 2008

Learning of Introduction Semantic Relations between Ontology Concepts using Statistical Techniques ◮ A methodology for automatic learning of ontologies A. Tegos from texts which are semantically annotated with Introduction instances of ontologies’ concepts The Proposed Method Finding the Semantic Relations of concepts ◮ Applying statistical techniques to metadata extracted Finding the Cardinality from the annotated texts we discover: Restrictions Experimental - semantic relations among the annotated concepts Assessment - cardinality restrictions for these relations Conclusions Future Plans ◮ The method was applied to corpora from two different domains, athletics and biomedical , and was evaluated against the existing manually created ontologies for these domains

Learning of Outline Semantic Relations between Ontology Concepts using Statistical Techniques A. Tegos Introduction Introduction The Proposed The Proposed Method Method Finding the Semantic Relations of concepts Finding the Semantic Relations of concepts Finding the Finding the Cardinality Restrictions Cardinality Restrictions Experimental Assessment Experimental Assessment Conclusions Future Plans Conclusions Future Plans

Learning of Basic assumption Semantic Relations between Ontology Concepts using Statistical Techniques Our method is based on the assumption that concepts which A. Tegos are semantically related, tend to be “near” as context in a plain text Introduction The Proposed ◮ This assumption arises from the principle of coherence Method Finding the Semantic on linguistics Relations of concepts Finding the Cardinality Restrictions The discovery process is not based to commonly used Experimental Assessment assumptions: Conclusions ◮ Verbs typically indicate semantic relations Future Plans ◮ Does not exploit lexico-syntactic patterns or clustering methods ◮ Does not use any external knowledge sources like WorldNet

Learning of Definitions Semantic Relations between Ontology Concepts using Statistical Techniques A. Tegos ◮ Low-Level : concepts whose instances are associated with relevant text portions Introduction e.g. name(has-instance) or the age(has-instance) The Proposed Method Finding the Semantic Relations of concepts ◮ High-Level : “compound” concepts in such a way that Finding the Cardinality Restrictions instances of these concepts are related to instances of Experimental low-level concepts Assessment Conclusions e.g. person(name, age, nationality, gender) Future Plans ◮ We focus on the discovery of semantic relations between high-level concepts, but we also show the applicability of the proposed approach to low-level concepts

Learning of Requirements Semantic Relations between Ontology Concepts using Statistical Techniques A. Tegos Introduction The Proposed ◮ The method requires the annotation of the corpus with Method Finding the Semantic instances of ontology’s concepts. Relations of concepts Finding the Cardinality Restrictions ◮ In the case of high-level concepts as instances we Experimental Assessment consider the fillers of the concept’s attributes that have Conclusions been found in a document. Future Plans

Learning of An example of the annotation Semantic Relations between Ontology Concepts using Statistical Techniques A. Tegos The 34-year-old, World marathon record holder and two-time Olympic Introduction and four-time World 10,000m champion Haile Gebreselassie of Ethiopia The Proposed Method today announced that he intends to compete in this 2008 FKB-Games - Finding the Semantic IAAF World Athletics Tour - in Hengelo, the Netherlands on 24 May in Relations of concepts Finding the his bid to make Ethiopia’s team for the Beijing Olympics in China. Cardinality Restrictions Experimental Assessment Athlete (name: Haile Gebreselassie , age: 34 , nationality: Ethiopia , Conclusions gender: NotFound ) Future Plans SportsCompetition (sport-name: 10,000m , city: Hengelo , stadium-name: NotFound , date: 24 May )

Learning of The proposed method Semantic Relations between Ontology Concepts using Statistical Techniques A. Tegos The proposed method for ontology learning involves 2 major Introduction steps: The Proposed Method Finding the Semantic Relations of concepts Finding the ◮ Finding the semantic relations of concepts that have Cardinality Restrictions been annotated in the corpus. Experimental Assessment Conclusions ◮ Finding the cardinality restrictions for the extracted Future Plans relations.

Learning of 1. Finding the offsets of the annotated instances Semantic Relations between Ontology Concepts using Statistical Techniques ◮ Based on our assumption, we treat each document of A. Tegos the corpus as a sequence of symbols. Introduction The Proposed Method ◮ In this manner, each document is represented in a Finding the Semantic Relations of concepts one-dimensional Euclidean space, depending on the Finding the Cardinality Restrictions place in which each symbol is found in the text. Experimental Assessment Conclusions ◮ We find for each document the offsets of the annotated Future Plans instances. ◮ As offset of an instance is defined the set that represents the minimum part of text which encloses all its fillers.

Learning of Example for the offset of the annotated instances Semantic Relations between Ontology Concepts using The 34-year-old, World marathon record holder and two-time Olympic and Statistical four-time World 10,000m champion Haile Gebreselassie of Ethiopia today Techniques announced that he intends to compete in this 2008 FKB-Games - IAAF World A. Tegos Athletics Tour - in Hengelo, the Netherlands on 24 May in his bid to make Ethiopia’s team for the Beijing Olympics in China. Introduction The Proposed Athlete (name: Haile Gebreselassie , age: 34 , nationality: Ethiopia , Method Finding the Semantic gender: NotFound ) Relations of concepts Finding the SportsCompetition (sport-name: 10,000m , city: Hengelo , stadium-name: Cardinality Restrictions NotFound , date: 24 May ) Experimental Assessment Conclusions ◮ The offset of the document is the set [0 , 342]. Future Plans ◮ The offset of the phrase “ 34-year-old, World marathon ” is the set [4 , 30] ◮ The offset for the Athlete ’s instance is the set [4 , 134]. ◮ The offset for the SportsCompetition ’s instance is the set [87 , 270]

Learning of 2. Finding overlapping instances Semantic Relations between Ontology Concepts using Statistical Techniques ◮ For each document, we search for the different pairs of A. Tegos concepts that have overlapping instances: Introduction The Proposed For the document doc z , of the corpus: Method C doc z = { C 1 , C 2 , . . . , C n } where C i = { I 1 , I 2 , . . . , I m } Finding the Semantic I k = [ l , r ] T N Relations of concepts where and l < r , Finding the Cardinality we compare the instances’ offsets: Restrictions ∀ ( I x , I y ) where I x ∈ C i , I y ∈ C j Experimental Assessment C i ∈ C doc z C j ∈ C doc z − { C i } and and Conclusions “ ” “ ” \ I y � = ∅ If I x then create a pair C i , C j for doc z (1) Future Plans ◮ Note that for each document we are interested only in finding the different pairs of related concepts and not the number of occurrences for each of these pairs.

Learning of 3. The semantic-correlation metric Semantic Relations between Ontology Concepts using Statistical Techniques ◮ This metric measures the tendency of concept C i to be A. Tegos semantically related, either taxonomically or Introduction non-taxonomically, with concept C j , but not the inverse. The Proposed Method Finding the Semantic „ « Relations of concepts S ( C i → C j ) = P ( C j | C i ) · 1 + I ( C i , C j ) = Finding the Cardinality Restrictions «! Experimental „ P ( C j | C i ) = P ( C j | C i ) · Assessment 1 + log (2) P ( C i ) · P ( C j ) Conclusions Future Plans ◮ This definition is based on our assumption that concepts which are semantically related, tend to co-occur “near”. Therefore, concepts whose instance offsets overlap frequently tend to be semantically related.