Contextual Semantic Integration Brendan Tierney School of Computing
Contextual Semantic Integration Brendan Tierney School of Computing, Dublin Institute of Technology, Kevin St. , Dublin 8, Ireland. © Brendan Tierney 2003/2004. brendan. tierney@dit. ie Doctorial Consortium British National Conference on Databases 6 th July, 2004.
Contents © Brendan Tierney 2003/2004. brendan. tierney@dit. ie § Overview of Research Topic § Problem Description § Current Ontology Integration Approaches § Proposed Solution § Similar Solutions § Work Plan 2
Overview of Research Topic § Ontology Integration § Existing approaches to integration still relies on human intelligence § Structural, syntactic, etc § Need some way to build in meaning § Contextual Semantic Integration § Why do we need to add meaning § Due to ontology versioning, evolution § Different localised interpretations § How can we add meaning § Using some form of dictionary § But words can have different meaning § Need to isolate the correct meaning i. e. give it the correct context § Context given by set of semantically equivalent words § Integration based on these sets © Brendan Tierney 2003/2004. brendan. tierney@dit. ie § Main application area being business intelligence and KDD § Better business integration § Better business understanding 3
Problem Description § Typical data integration Sales App 4 Marketing © Brendan Tierney 2003/2004. brendan. tierney@dit. ie App 5 CRM 4
Problem Description § Need to work towards CRM App 4 App 5 Global Model Marketing Sales © Brendan Tierney 2003/2004. brendan. tierney@dit. ie 'The Involvement of Human Resource in Large Scale Data Mining Projects 5 Hoffmann & Tierney, 2003, Symposium on Information and Communication Technologies
Problem Description § Problems § § Ad-hoc programs Need large amount of human involvement High maintenance cost High probability of errors being introduced § Example - O 2 World. View § § § § © Brendan Tierney 2003/2004. brendan. tierney@dit. ie § For top 1000 customers Give a complete view of mobile phone costs Each country had different applications World View model created § Agreed by customers § Agreed by countries Each country had to supply data using World View model Each country had different meanings for terms used in World View model Became a much larger project § Detailed definitions needed to be worked out and agreed § Needed to re-negotiate definition several times § No clear meanings from source applications Global Ontology / Corporate Ontology / Local Ontology § All had different meanings based on environment 6
Current Ontology Integration Approaches Architectures Programme Committee Member Workshop on Knowledge Discovery and Ontologies © Brendan Tierney (KDO-2004) 2003/2004. ECML/PKDD 2004 – Italy – brendan. tierney@dit. ie Tier Architect ure 'Using Ontologies in Knowledge Discovery in 7 Data'
Current Ontology Integration Approaches § Syntactic § Only concerned with the content of the data § The structure or semantics is not considered § Matches concepts using syntax driven techniques and syntactic similarity measures § Similarities based on attribute/label correspondence and data similarity matching on sub-strings § (“phone”, “telephone”), soundex and abbreviations (“Street”, “St”) § Only suitable in a small number of cases § Understanding the content and assessing its value for a given task requires human intervention (a lot) § Structural § Typically performed by ad-hoc programs § Requires domain specific expertise § Structure and position of the node within the graphical representation of the ontology § Involves § Exact § Inexact or approximate matching © Brendan Tierney 2003/2004. brendan. tierney@dit. ie § Semantic § Most approaches involve semantic relationships, homonyms and synonyms § Real meaning of concept is not available or used 8
Proposed Solution § Contextual Semantic Integration (CSI) § An Ontology is a model of some domain which is supposed to encode a view common to a set of different parties An ontology is built to be shared; § A Context is a model of some domain which is supposed to encode a view of a party A context is built to be kept local (where local implies not shared) § The Need for Context § No common set of ontologies § Ontology versioning and evolution causes changes § Regional and inter-organisational semantic differences § Causing semantic conflicts § Current ontology approaches do not facilitate meaning of concepts § Semantics are only via relationships, etc © Brendan Tierney 2003/2004. brendan. tierney@dit. ie § The approach to providing meaning (with a KDD in mind) § Set of semantically equivalent words (SEW) § From some available dictionary/thesaurus (maybe Word. Net) 'Contextual for Ontology § Expand current ontology language to facilitate this. Extensions (maybe OWL) 9 Integration'
Proposed Solution § Simple Example Name name >> <<name Education <<education> Work <work> < work> © Brendan Tierney 2003/2004. brendan. tierney@dit. ie CV >> <<CV CV Private <private> < private > 10
Proposed Solution § CSI Approach § Ontologies are compared for commonality in list of SEWs § Comparison of concepts § Possibilities § 2 concepts have the same name and the same set of SEWs § They have the same contextual meaning § Are equivalent § 2 concepts have same name but different set of SEWs § They do not have the same contextual meaning § Are not equivalent § 2 concepts (name not important) but similar set of SEWs § Depends on the degree of similarity of the set of SEWs § Need some way to measure this © Brendan Tierney 2003/2004. brendan. tierney@dit. ie 11
Proposed Solution § Example § Room (O 1) § From Word. Net § (527) room -- (an area within a building enclosed by walls and floor and ceiling; "the rooms were very small but they had a nice view") § SEWs = <Space, Walls, Floor, Ceiling, Window, Door, Building> § Room (O 2) § From Word. Net § (36) room, way, elbow room -- (space for movement; "room to pass"; "make way for"; "hardly enough elbow room to turn around") § SEWs = <Space, Movement, Elbow, Distance, Spatial> § Different contextual meaning © Brendan Tierney 2003/2004. brendan. tierney@dit. ie § Need to allow negation to allow a greater degree of meaning 12
Proposed Solution § Architecture § Contextual Selector § Input = Ontology § Output = Ontology + SEW (created using Word. Net) § Contextual Integration § In conjunction with syntactic and structural integration § Contextual Semantic Distance § Used to measure the degree of similarity between concepts § Involves measuring at the concept level rather than at ontology level § Additional research needed to compare with similarity measures in IR © Brendan Tierney 2003/2004. brendan. tierney@dit. ie The appropriate setting for 13 the Threshold will be investigated during the
Similar Solutions § C-OWL : Contextualizing Ontologies (Bouquet et al 2003) § Contextual Ontology = Ontology + Context Mappings § Is a local ontology plus a set of bridge rules Contextual Ontology = Ontology + § Built using mappings between ontologies SEWs § Contextual semantics is built into the mappings § No the ontology or version of it § Still don’t know the meaning of the concepts. § Momis (Beneventano, 2003) Generates a Global Virtual View (GVV) Uses the local views with a thesaurus (uses Word. Net) Local views have a common domain Semantics determined by definition in local view and names in thesaurus Use Word. Net to obtain § Real meaning of the concepts is not considered meaning and to build up SEW § § § Others include © Brendan Tierney 2003/2004. brendan. tierney@dit. ie § COG, Info. Sleuth, PROMPT, ONIONS, Unicorn 14 My approach does
Work Plan § March 2004 – Sept 2004 § § § Complete first draft literature review Complete problem description Submit paper and present at BNCOD Initial specification of semantic integration using context § § Journal paper on background and key issues (Journal on Information & Software Technology Detailed specification of semantic integration using context § § Use Word. Net or some other approach for providing semantics Create & Setup test ontologies Paper on proposed solution using Word. Net or some other approach Write up dissertation for first draft of solution © Brendan Tierney 2003/2004. brendan. tierney@dit. ie Usingle words for labels Jan. 2006 – June. 2006 § § § (ish) Sept. 2004 – Dec. 2005 § § Specification of context integration based multi-words Simulations of multi-word approach Evaluations of single vs multi-word approaches Paper based on evaluations Write up dissertation for second draft of solution July. 2006 – Dec. 2006 § § § Review lit review to include resent research developments Discussion of benefits and weaknesses Write up evaluation section of dissertation Complete write up of dissertation Submit dissertation for examination 15
- Slides: 15