Geo-ontology Integration: Identifying Issues, Dimensions and Developing Guidelines M. Kavouras & M. Kokla School of Rural and Surveying Engineering National Technical University of Athens Joint ISPRS WG II/3 - WG II/6 Workshop on Multiple Representation and Interoperability of Spatial Data, Hannover, Germany, February, , 2006

Structure of the presentation Definition of fundamental terms Different perspectives Principal characteristics of ontology integration Fundamental questions of ontology integration Three sub-processes Directions to ontology integration

What is meant by “ontology” Ontology is usually defined as “a formal, explicit specification of a shared conceptualization of a domain” (Gruber, 1993). Provide complete and commonly accepted descriptions – documentations of the concepts of a domain GEOGRAPHIC ONTOLOGIES are usually terminological: a concept is described by a term, a natural language definition and relations to other concepts.

What is meant by “integration” Different terms to denote a number of related processes: association, coordination, combining, matching, mapping, translation, merging, alignment, unification, etc. Fundamental objective of all approaches: 1.Compare the semantics of original ontologies 2.Determine the following: Whether the given ontologies are to some degree similar, related, or disjoint. How to compare concepts in order to identify similarities and heterogeneities. How to associate the original ontologies on the basis of the previous findings.

What is meant by “ontology integration” Ontology integration: integration of different existing ontologies (inter-ontology mapping) vs. Ontology-based integration: integration of different database schemata to a single reference (top- level) ontology vs. Data integration: based on a single model.

Two perspectives A “higher” ontological perspective with an interest in conceptualizing and representing knowledge about a domain (in our case, geographic reality)  semantic conflicts due to different conceptualizations and models of the domain in an information system. A “lower” explication perspective with an interest in formalizing, processing and associating existing information or data.  conflicts in the specification of the conceptualization (e.g., encoding differences, representation language mismatches).  terminological conflicts can be treated at the explication level; but they often carry some semantic weight

Ontological and semantic notions are used differently according to TWO PERSPECTIVES A “higher” ontological perspective A “lower” design/implementation perspective conceptualization differences explication differences terminological differences

Conceptualization Differences (1) Perspective/interest. Often what determines the concepts and taxonomies to be designed or adopted is the application needs (different application needs create different taxonomies). Disciplinary training. Disciplines tend to develop a common understanding of their domain knowledge. Methodology. In the scientific context, the methods we employ often determine to a great extend what is we see and how we partition reality (e.g., land cover nomenclatures according to the interpretation method used - remote sensing).

Conceptualization Differences (2) Granularity. The scale of analysis determines not only the taxonomical detail but may create completely different taxonomies (e.g., a 1: land cover nomenclature differs considerably from that of 1: 5000). Ethno-/cultural-/socio-based view. Many geographic concepts of a domain are the result of constructive social agreement and partial consensus. Human cognitive diversity. When people work autonomously, they perceive and conceptualize geospace differently, creating thus their own cognitive taxonomies.

Taxonomic ≠ diversity DISCIPLINARY TRAINING COGNITIVE DIVERSITY INTEREST/ PERSPECTIVE ETHNO-/SOCIO- /CULTURAL-VIEW GRANULARITY METHODOLOGY

Principal characteristics of ontology integration 1.Assumptions made about the source of semantics and the objective of the process 2.Semantic level addressed 3.Input (source) / Output components 4.Method used 5.Degree of change - alteration caused to the original ontologies 6.Degree of interaction or user involvement

Fundamental questions of ontology integration (1) Q1: Which semantics affect integration? Action: Define the semantic elements which shall prevail integration. Q2: Where semantics emanate from? Action: Determine the available sources of semantics. Q3: How is semantics derived? Action: Use appropriate approach to extract semantic components from available sources. Q4: How are concepts and ontologies compared? Action: Define which concepts and ontologies are to be compared and the basis of their comparison. Q5: How is similarity/heterogeneity among concepts determined? Action: Decide on how the comparison takes place and what the possible/acceptable outcome of the comparison is.

Fundamental questions of ontology integration (2) Q6: How is heterogeneity resolved/reconciled? Action: Decide on how heterogeneities among concepts are reconciled Q7: What type of integration is preferred? Action: Define what resource ontologies are to be integrated and their role during and after integration, if a target ontology is used to guide integration, etc. Q8: Is user/expert involvement essential in the process? Action: Determine the degree of automation or interaction/expert involvement needed in resolving complex cases. Q9: How is the result evaluated? Action: Determine the basis of assessing the result of integration, how objective/subjective the result may be, what kind of inconsistencies are expected, should be avoided, etc.

Three sub- processes for semantic integration ΕΞΑΓΩΓΗ ΣΗΜΑΣΙΟΛΟΓΙΚΗΣ ΠΛΗΡΟΦΟΡΙΑΣ SEMANTIC INFORMATION EXTRACTION ΟΛΟΚΛΗΡΩΣΗ INTEGRATION Αρχικές έννοιες-Οντολογίες Input concepts - ontologies - Σημασιολογικά πλούσιες έννοιες Semantically rich concepts Ομοιότητες και ετερογένειες Similarities and heterogeneities Ολοκληρωμένη οντολογία Integrated ontology Διαδικασία 3Process III Διαδικασία 1Process I Διαδικασία 2Process II ΣΥΓΚΡΙΣΗ ΚΑΤΗΓΟΡΙΩΝ CONCEPT COMPARISON

Process I: Semantic Information Extraction Source components: free text, corpora, thesauri, specialized text (e.g., definitions), terms, nomenclatures, data dictionaries, hierarchical classifications, database schemata, etc. Taxonomic ontologies vs. formal ontologies. What constitutes semantic information ? Empirical ad hoc approaches attempting to formalize the concepts involved, and design the associated databases. Information extraction (IE) approaches based on NLU/NLP - central terms in computational linguistics and artificial intelligence.

Process II: Concept - Ontology Comparison Comparison and similarity measures reveal/depict how difficult integration (Process III) will be. A comparison shall reveal and somehow measure similarities or heterogeneities (conflicts). Similarity between geographic concepts can be estimated by combining feature and linguistic matching, and semantic distance calculation (Tversky, 1977; Rodríguez & Egehhofer 2002; Yaolin et al. 2002). Process II also needs to resolve the heterogeneities.

Process III: Integration (1) Alignment is a mapping between concepts of different ontologies bringing them into mutual agreement.  Translation/conversion utilities are used to provide functionality.  No ontology is distorted.  A target ontology may or may not be aligned with the resource ontologies. Partial compatibility creates a merging of only those parts of ontologies that are considered more similar.  The merged parts distort the initial common ontology parts.  A target ontology may or may not be used for the merging of the common parts.

Process III: Integration (2) Unification (also fusion), extends partial compatibility to all ontologies and their concepts.  Each resource ontology is distorted to become fully compatible with the others; there is a single ontology at the end.  The initial ontologies are distorted.  A target ontology may or may not be used for defining the unified ontology. True integration creates a single integrated ontology whose parts are the resource ontologies including some additional concepts necessary for the association.  The user deals with a single integrated ontology.  The resource ontologies are not distorted  A target ontology may or may not be used in the integration.

m1 m2 m3 m4 m Alignment Partial compatibility Unification True integration

Principal directions to ontology integration 1.Conforming to a single central ontology 2.Manual ad-hoc mappings 3.Intuitive mappings based on “light” lexical information 4.Intuitive mappings based on explication characteristics 5.Intuitive mappings based on structural similarity 6.Relating (grounding) to a single shared or top-level ontology 7.Direct mappings based on “deep” semantics 8.Integration by view-based query processing 9.Compound similarity measures 10.Extensional mappings based on common spatial reference

TO DISCUSS Core ideas behind intelligent integration Geospatial semantics (properties…) What is missing in the framework? Identify research “holes”

DISCUSSION 2 What is special about SPATIAL ONTOLOGIES Vocabulary – ontology of ontologies ONTOLOGIES (KR)- YES, SEMANTICS- NO (…later) 1. HOW to go from Simple iconic level to the formal level 2. From the formal level to the semantic level 3. Establish properies – context (neighborhood)  SEMANTIC IS NOT THEMATIC

DISCUSSION 3 Cognitive basis for categorization Research on ontologies needed Not real ontologies existing Work on combining extensional and intensional information useful - complementary Difference between schema and ontology integration  Missing : contextual analysis, relationships

DISCUSSION 3  Forest ?????  Similar concepts may have different reps TOP LEVEL ONTOLOGIES? Ambitious goal – USE the same ontology to compare different domains How to derive semantic properties What is special about SPATIAL