Dipso K Mi


Published on

Published in: Education, Technology
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide
  • The rest of the talk focuses on specific results we have obtained in the past 12 months, so I won’t really spend any time on discussing this NGSW paradigm in any detail. If you guys are interested in finding out more, we published the ‘definitive paper’ a few months ago, which describes the vision, relation to the evolution of AI, tech infrastructure, and concrete technologies;
  • Dipso K Mi

    1. 1. Web3.0 and Language Resources Knowledge Media Institute (KMi) The Open University Semantic Technologies @ KMi
    2. 2. Outline <ul><li>Library of generic problem solving methods </li></ul><ul><ul><li>To act as active components </li></ul></ul><ul><li>Methods for dealing with heterogeneous knowledge sources </li></ul><ul><ul><li>Knowledge Fusion (KnoFuss) </li></ul></ul><ul><ul><li>Ontology Matching (Scarlet) </li></ul></ul><ul><li>Infrastructures for storing large scale semantics (Watson) </li></ul><ul><li>Tools for exploiting distributed semantics </li></ul><ul><ul><li>Open Domain, Multi-Ontology Question Answering (PoweAqua) </li></ul></ul>
    3. 3. PSM <ul><li>Library of generic problem solving methods </li></ul><ul><ul><li>To act as active components </li></ul></ul><ul><ul><li>I include both generic slides and a detailed example- this is more for you to understand what is going on, I think most of the detailed slides can be skipped in the talk. </li></ul></ul>
    4. 4. Knowledge-level Architectures for Sharing and Reuse Application of the modelling paradigm to the specification and use of libraries of reusable components for knowledge systems Knowledge-level Architectures for Sharing and Reuse
    5. 5. Modelling Frameworks (1) <ul><li>A modelling framework identifies the generic types of knowledge which occur in knowledge systems, thus providing a generic epistemological organization for knowledge systems </li></ul><ul><li>Several exist </li></ul><ul><ul><li>KADS/Common KADS - Un.of Amsterdam </li></ul></ul><ul><ul><li>Components of Expertise - Steels </li></ul></ul><ul><ul><li>Generic Tasks - Chandrasekaran </li></ul></ul><ul><ul><li>Role-limiting Methods - McDermott </li></ul></ul><ul><ul><li>Protégé - Musen, Stanford </li></ul></ul><ul><ul><li>TMDA - Motta </li></ul></ul><ul><ul><li>Ibrow - Fensel, Motta et al. </li></ul></ul>
    6. 6. Modelling Frameworks (2) <ul><li>Much in common </li></ul><ul><ul><li>Emphasis on reusable models </li></ul></ul><ul><ul><li>Typology of generic tasks </li></ul></ul><ul><ul><li>Constructivist paradigm </li></ul></ul><ul><li>Some differences </li></ul><ul><ul><li>Different degrees of coupling between domain-specific and domain-independent knowledge </li></ul></ul><ul><ul><li>Different degrees of flexibility </li></ul></ul><ul><ul><li>Different typologies of knowledge categories </li></ul></ul>
    7. 7. A Constructive Approach... Let’s define our own framework...
    8. 8. Generic Tasks <ul><li>Informal definition </li></ul><ul><ul><li>A generic class of applications - e.g., planning, design, diagnosis, scheduling, </li></ul></ul><ul><li>More precise definition </li></ul><ul><ul><li>A knowledge-level, application-independent description of the goal to be attained by a problem solver. </li></ul></ul><ul><li>Several typologies exist </li></ul><ul><ul><li>e.g., Breuker, 1994 </li></ul></ul><ul><li>Viewpoints over applications </li></ul><ul><ul><li>No ‘natural categories’ </li></ul></ul><ul><ul><li>Different viewpoints can be imposed on a particular application </li></ul></ul>
    9. 9. Example: Parametric Design <ul><li>Generic Task Parametric Design </li></ul><ul><li>Inputs : Parameters, Constraints, Requirements, Cost-Function, Preferences </li></ul><ul><li>Output : Design-Model </li></ul><ul><li>Goal : “To produce a complete and consistent design model, which satisfies the given requirements” </li></ul><ul><li>Preconditions : “At least one requirement and one parameter are provided” </li></ul>
    10. 10. Example: Classification <ul><li>Generic Task Classification Inputs : Candidate-classes Observables Match-criterion Solution-criterion </li></ul><ul><li>Output : Best-Matching-Classes </li></ul><ul><li>Preconditions : “At least one candidate class exists” </li></ul><ul><li>Goal : “To find the class that best explains the observables” </li></ul>
    11. 11. Generic Component 2: Reusable PSMs <ul><li>A domain-independent, knowledge-level specification of problem solving behaviour, which can be used to solve a class of tasks. </li></ul><ul><li>PSM specifications may be partial </li></ul><ul><li>PSM can be task-specific </li></ul><ul><ul><li>E.g., heuristic classification </li></ul></ul><ul><li>PSM can be task-independent </li></ul><ul><ul><li>E.g., search methods, such as hill-climbing, A*, etc..... </li></ul></ul>
    12. 12. Functional Specification of a PSM <ul><li>Problem solving method search </li></ul><ul><li>ontology </li></ul><ul><li>import </li></ul><ul><li>state-space-terminology </li></ul><ul><li>competence </li></ul><ul><li>roles </li></ul><ul><li>input input: State </li></ul><ul><li>output output: State </li></ul><ul><li>preconditions </li></ul><ul><li>input ≠ 0 </li></ul><ul><li>postconditions </li></ul><ul><li>solution_state (output) </li></ul><ul><li>assumptions </li></ul><ul><li> ?s . solution_state (?s) & successor (input, ?s) </li></ul>
    13. 13. Operational Description <ul><li>Begin </li></ul><ul><li>states:= one x. initialize ( input input) </li></ul><ul><li>repeat </li></ul><ul><li>state:= one x . select _state ( states states) </li></ul><ul><li>if solution_state (state) </li></ul><ul><li>then return state </li></ul><ul><li> else </li></ul><ul><li> succ_states:= one x. derive_successor_states ( state state) </li></ul><ul><li> states:= one x. update_state_space ( input1 states input2 succ_states) </li></ul><ul><li>end if </li></ul><ul><li>end repeat </li></ul><ul><li>end </li></ul>
    14. 14. Task-Method Structures Problem Type Primitive PSM
    15. 15. Multi-Functional Domain Models <ul><li>Domain-specific models, which are not committed to a specific PSM or task. </li></ul><ul><li>Examples </li></ul><ul><ul><li>A database of cars </li></ul></ul><ul><ul><li>The CYC knowledge base, etc.. </li></ul></ul>
    16. 16. Picture so far.. Problem Solving Method Classification Simple Classifier Lunar rocks Application Model Generic Task Multi-Functional Domain
    17. 17. Issue <ul><li>How to link different reusable components? </li></ul>Problem Solving Method Classification Simple Classifier Lunar rocks Application Model Generic Task Multi-Functional Domain
    18. 18. Solution: Mappings <ul><li>Mappings model explicitly the relationship between different components in an application model </li></ul>Problem Solving Method Classification Task-Domain Mapping PSM-Domain Mapping Simple Classifier Lunar rocks Application Model Generic Task Multi-Functional Domain Task-PSM Mapping
    19. 19. Example <ul><li>Scenario: Office Allocation Application </li></ul><ul><li>Generic Task: Parametric Design </li></ul><ul><li>Domain: KB about employees and offices </li></ul>Parameter Employee Design Model Pairs <Employee, Room> Task Level Domain Level
    20. 20. <ul><li>Mappings are an example of application-specific knowledge. Are there others? </li></ul>Application-specific knowledge Yes: Application-specific heuristic problem solving knowledge
    21. 21. Elevator Design Example <ul><li>A configuration designer only considers two positions for the counterweight </li></ul><ul><ul><li>Half way between platform and U-bracket </li></ul></ul><ul><ul><li>A position such that the distance between the counterweight and the platform is at least 0.75 inches </li></ul></ul>
    22. 22. Complete Picture Problem Solving Method Application Model Generic Task Multi-Functional Domain Mapping Knowledge Application-specific Problem-Solving Knowledge Application Configuration
    23. 23. Detailed Example: A Library of Components for Classification
    24. 24. Classification <ul><li>Classification can be seen as the problem of finding the solution (class), which best explains a set of known facts (observables), according to some criterion </li></ul>Observables Candidate Sols. Criterion Classification Solution
    25. 25. Example Observables Candidate Sols. Criterion Classification Solution {background=green; area=china...} Complete-coverage-criterion (every observable has to be explained) {chinese-granny, dutch-granny, etc..} {chinese-granny}
    26. 26. Observables <ul><li>Observables = set_of (Observable); </li></ul><ul><li>Observable = {feature, value}. </li></ul><ul><li>Well defined Observables (obs): </li></ul><ul><li>({f 1 , v 1 }  obs  {f 1 , v 2 }  obs) -> v 1 = v 2 </li></ul><ul><li>({f 1 , v 1 }  obs) -> legal_feature_value (f 1 , v 1 ) </li></ul>
    27. 27. Solutions <ul><li>Solution = set_of (Feature_Spec); </li></ul><ul><li>Feature_Spec = {Feature, Feature_value_spec} </li></ul><ul><li>Feature_value_spec = Unary_Relation </li></ul><ul><li>Well defined Solution (sol): </li></ul><ul><li>{f 1 , s 1 }  sol  holds (s 1 , v 1 ) -> legal_feature_value (f 1 , v 1 ) </li></ul>
    28. 28. Matching <ul><li>Observable={f 1 , v 1 } matches Solution=sol iff : </li></ul><ul><li>{f 1 , c}  sol  holds (c, v 1 ) </li></ul>
    29. 29. Matching Sets of Obs to a Solution <ul><li>Sol: {{fsol 1 , c 1 }...{fsol m , c m }}; Obs: {{fob 1 , v 1 }...{fob n , v n }} </li></ul><ul><li>Four possible cases: </li></ul><ul><li>{f j , c j }  sol  {f j , v j }  obs  holds (c j , v j ) -> Explained (f j ) </li></ul><ul><li>{f j , c j }  sol  {f j , v j }  obs  not holds (c j , v j ) -> Inconsistent (f j ) </li></ul><ul><li>{f j , v j }  obs  {f j , c j }  sol -> Unexplained (f j ) </li></ul><ul><li>{f j , v j }  obs  {f j , c j }  sol -> Missing (f j ) </li></ul>
    30. 30. Default Match Criterion <ul><li>Match Score : </li></ul><ul><ul><li>Vector: <I, E, U, M> </li></ul></ul><ul><li>Match Comparison Relation </li></ul><ul><li>S 1 = (i 1 , e 1 , u 1 , m 1 ); S 2 = (i 2 , e 2 , u 2 , m 2 ) </li></ul><ul><li>S 1 better_score than S 2 iff: </li></ul><ul><li>(i 1 < i 2 )  </li></ul><ul><li>(i 2 = i 1  e 2 < e 1 )  </li></ul><ul><li>(i 2 = i 1  e 2 = e 1  u 1 < u 2 )  </li></ul><ul><li>(i 2 = i 1  e 2 = e 1  u 2 = u 1  m 1 < m 2 ) </li></ul>
    31. 31. Possible Solution Criteria <ul><li>Positive Coverage </li></ul><ul><ul><li>Some feature is explained and none is inconsistent </li></ul></ul><ul><li>Complete Coverage </li></ul><ul><ul><li>All features are explained and none is inconsistent </li></ul></ul>
    32. 32. Hierarchy of Criteria Match Criterion Match Score Comparison Rel Macro Score Mechanism Feature Score Mechanism Match Score Mechanism Solution Criterion
    33. 33. Observables <ul><li>(def-class observables (set) ?obs </li></ul><ul><li>&quot;This is simply a set of observables. </li></ul><ul><li>An important constraint is that there cannot be two values for the same feature </li></ul><ul><li>in a set of observables&quot; </li></ul><ul><li>:iff-def (every ?obs observable) </li></ul><ul><li>:constraint (not (exists (?ob1 ?ob2) </li></ul><ul><li>(and (member ?ob1 ?obs) </li></ul><ul><li>(member ?ob2 ?obs) </li></ul><ul><li>(has-observable-feature ?ob1 ?f) </li></ul><ul><li>(has-observable-feature ?ob2 ?f) </li></ul><ul><li>(has-observable-value ?ob1 ?v1) </li></ul><ul><li>(has-observable-value ?ob2 ?v2) </li></ul><ul><li>(not (= ?v1 ?v2)))))) </li></ul>
    34. 34. Solutions <ul><li>(def-class solution () ?x </li></ul><ul><li>&quot;A solution is a set of feature definitions&quot; </li></ul><ul><li>:iff-def (every ?x feature-definition)) </li></ul><ul><li>(def-class feature-definition () ?x </li></ul><ul><li>((has-feature-name :type feature) </li></ul><ul><li>(has-feature-value-spec :type unary-relation)) </li></ul><ul><li>:constraint (=> (and (has-feature-name ?x ?f) </li></ul><ul><li>(has-feature-value-spec ?x ?spec)) </li></ul><ul><li>(=> (holds ?spec ?v) </li></ul><ul><li>(legal-feature-value ?f ?v)))) </li></ul>
    35. 35. Solution Criterion <ul><li>(def-class solution-admissibility-criterion () ?c </li></ul><ul><li>((applies-to-match-score-type :type match-score-type) </li></ul><ul><li>(has-solution-admissibility-relation :type unary-relation)) </li></ul><ul><li>:constraint (=> (and (solution-admissibility-criterion ?c) </li></ul><ul><li>(has-solution-admissibility-relation ?c ?r) </li></ul><ul><li>(domain ?r ?d)) </li></ul><ul><li>(subclass-of ?d match-score))) </li></ul>
    36. 36. Monotonicity of Admissibile Solutions <ul><li>(def-axiom admissibility-is-monotonic </li></ul><ul><li>&quot;This axiom states that the admissibility criterion is monotonic. That is, if a solution, ?sol, is admissible, then any solution which is better than ?sol will also be admissible&quot; </li></ul><ul><li>(forall (?sol1 ?sol2 ?obs ?criterion) </li></ul><ul><li>(=> (and (admissible-solution </li></ul><ul><li>?sol1 (apply-match-criterion ?criterion ?obs ?sol1) ?criterion) </li></ul><ul><li>(better-match-than ?sol2 ?sol1 ?obs ?criterion)) </li></ul><ul><li>(admissible-solution </li></ul><ul><li>?sol2 (apply-match-criterion ?criterion ?obs ?sol2) ?criterion)))) </li></ul>
    37. 37. Complete Coverage <ul><li>(def-instance complete-coverage-admissibility-criterion </li></ul><ul><li>solution-admissibility-criterion </li></ul><ul><li>((applies-to-match-score-type default-match-score) </li></ul><ul><li>(has-solution-admissibility-relation </li></ul><ul><li>complete-coverage-admissibility-relation))) </li></ul><ul><li>(def-relation complete-coverage-admissibility-relation (?score) </li></ul><ul><li>&quot;a solution should be consistent and explain all features&quot; </li></ul><ul><li>:constraint (default-match-score ?score) </li></ul><ul><li>:iff-def (and (= (length (first ?score)) 0) ;; no inconsistency </li></ul><ul><li>(= (length (third ?score)) 0))) ;; no unexplained </li></ul>
    38. 38. Classification Task Ontology <ul><li>42 Definitions </li></ul><ul><li>Provides both a theory of classification and a vocabulary to describe classification problems </li></ul><ul><li>Ontology is separated from task specifications </li></ul>
    39. 39. Generic Classification Task <ul><li>Input roles </li></ul><ul><ul><li>Candidate Solutions, Match Criterion, Solution Criterion, Observables </li></ul></ul><ul><li>Precondition </li></ul><ul><ul><li>Both observables and candidate solutions have to be provided </li></ul></ul><ul><li>Goal </li></ul><ul><ul><li>To find a solution from the candidate solutions which is admissible with respect to the given observables, solution criterion and match criterion </li></ul></ul>
    40. 40. Specific Classification Tasks <ul><li>Single-Solution Classification Task </li></ul><ul><ul><li>Single-solution assumption </li></ul></ul><ul><li>Optimal Classification Tasks </li></ul><ul><ul><li>Goal requires optimality </li></ul></ul>
    41. 41. Problem Solving Library <ul><li>Based on heuristic classification model </li></ul><ul><li>Supports both data-directed and solution-directed classification </li></ul><ul><li>Based on search paradigm </li></ul><ul><li>Supported by a method ontology </li></ul>
    42. 42. Method Ontology: Main Concepts <ul><li>Abstractors </li></ul><ul><ul><li>Mechanism for performing abstraction on observables </li></ul></ul><ul><ul><li>Abstractor: Obs* -> Obs </li></ul></ul><ul><li>Refiners </li></ul><ul><ul><li>Mechanism for specialising a solution </li></ul></ul><ul><ul><li>Refiner: Sol -> Sol* </li></ul></ul><ul><li>Candidate Exclusion Criterion </li></ul><ul><ul><li>A criterion which is used to decide when a search path is a dead-end </li></ul></ul><ul><ul><li>Default criterion rules out inconsistent solutions </li></ul></ul>
    43. 43. Monotonicity of Exclusion Criterion <ul><li>(def-axiom exclusion-is-monotonic </li></ul><ul><li>(forall (?sol1 ?sol2 ?obs ?criterion) </li></ul><ul><li>(=> (and (ruled-out-solution </li></ul><ul><li>?sol1 (the-match-score ?sol1) ?criterion) </li></ul><ul><li>(not (better-match-than ?sol2 ?sol1 ?obs </li></ul><ul><li>?criterion))) </li></ul><ul><li>(ruled-out-solution </li></ul><ul><li>?sol2 (the-match-score ?sol2)?criterion)))) </li></ul>
    44. 44. Axiom of Congruence <ul><li>(def-axiom congruent-admissibility-and-exclusion-criteria </li></ul><ul><li>(forall (?sol ?task) </li></ul><ul><li>(=> (member ?sol (the-solution-space ?task)) </li></ul><ul><li>(not (and (admissible-solution </li></ul><ul><li>?sol </li></ul><ul><li>(the-match-score ?sol) </li></ul><ul><li>(role-value </li></ul><ul><li>?task </li></ul><ul><li>'has-solution-admissibility-criterion)) </li></ul><ul><li>(ruled-out-solution </li></ul><ul><li>?sol </li></ul><ul><li>(the-match-score ?sol) </li></ul><ul><li>(role-value </li></ul><ul><li>?psm 'has-solution-exclusion-criterion))))))) </li></ul>
    45. 45. Three Heuristic Classification PSMs <ul><li>Two Data-directed </li></ul><ul><ul><li>Admissible Solution Classifier </li></ul></ul><ul><ul><ul><li>Finds one admissible solution according to the given criteria </li></ul></ul></ul><ul><ul><ul><li>Uses backtracking hill climbing </li></ul></ul></ul><ul><ul><li>Optimal Classifier </li></ul></ul><ul><ul><ul><li>Performs complete search looking for optimal solution </li></ul></ul></ul><ul><ul><ul><li>Uses best-first strategy </li></ul></ul></ul><ul><ul><ul><li>Uses candidate exclusion criterion to prune search space </li></ul></ul></ul><ul><li>One Solution-directed </li></ul><ul><ul><li>Goes down the solution hierarchy, acquiring observables as needed </li></ul></ul><ul><ul><li>Ask for observables with max discrimination power </li></ul></ul>
    46. 46. Task-Method Hierarchy
    47. 47. KnoFuss <ul><li>Methods for dealing with heterogeneous knowledge sources </li></ul><ul><ul><li>Knowledge Fusion (KnoFuss) </li></ul></ul><ul><li>The story here is that smart products will contain a lot of instance level semantic data encoded in terms of different ontologies and it will be important to be able to compare and merge similar (or fuse) instance data. </li></ul>
    48. 48. Knowledge fusion scenario RDF Images Other data Annotation Fusion Text Internal corporate reports (Intranet) Pre-defined public sources (WWW) Domain ontology KnoFuss Knowledge base
    49. 49. Fusion workflow Source KB Target KB SPARQL query translation Knowledge fusion Ontology integration Knowledge base integration Ontology matching Instance transformation Coreference resolution Dependency processing
    50. 50. KnoFuss architecture <ul><li>Method library </li></ul><ul><ul><li>Contains implementation of each specific algorithm </li></ul></ul><ul><li>Fusion ontology </li></ul><ul><ul><li>Describes method capabilities </li></ul></ul><ul><ul><li>Defines intermediate structures (mappings, conflict sets, etc.) </li></ul></ul>Fusion KB Intermediate data Main KB Fusion module ObjectIdentificationMethod ConflictDetectionMethod ConflictResolutionMethod Method library New data Fusion ontology
    51. 51. Steps <ul><li>Coreference resolution </li></ul><ul><ul><li>Attribute similarity algorithms </li></ul></ul><ul><li>Dependency processing </li></ul><ul><ul><li>Employing additional information: </li></ul></ul><ul><ul><ul><li>Schema restrictions </li></ul></ul></ul><ul><ul><ul><li>Links between instances </li></ul></ul></ul><ul><ul><ul><li>Provenance </li></ul></ul></ul><ul><ul><li>Using formal uncertainty reasoning </li></ul></ul><ul><ul><ul><li>Dempster-Shafer belief networks </li></ul></ul></ul>
    52. 52. Scarlet <ul><li>Methods for dealing with heterogeneous knowledge sources </li></ul><ul><ul><li>Ontology Matching (Scarlet) </li></ul></ul><ul><li>The motivation is that we will need to match between the different ontologies used by different products. </li></ul>
    53. 53. Ontology Matching 1 0.9 0.9 0.9 1 0.5 0.5 <ul><li>Label similarity methods </li></ul><ul><ul><li>e.g., Full_Professor = FullProfessor </li></ul></ul><ul><li>Structure similarity methods </li></ul><ul><ul><li>Using taxonomic/property related information </li></ul></ul>
    54. 54. Ontology Matching <ul><li>Most ontology matching techniques work only in cases when: </li></ul><ul><ul><li>There is a sufficient syntactic overlap between the labels of the concepts in the matched ontologies </li></ul></ul><ul><ul><li>The structure of the matched ontologies is rich enough to allow meaningful comparisons </li></ul></ul><ul><li>However, this might not be the case for smart products </li></ul><ul><ul><li>Smart products from different domains will use very different terminology thus excluding syntactic comparison </li></ul></ul><ul><ul><li>Smart product ontologies will probably be small and structurally shallow due to their resource limitations thus excluding the use of structure based techniques </li></ul></ul><ul><li>Therefore we propose a new matching paradigm which relies on the use of external knowledge </li></ul>
    55. 55. New paradigm: use of background knowledge A B Background Knowledge (external source) A’ B’ R R
    56. 56. External Source = Semantic Web <ul><li>Proposal: </li></ul><ul><li>rely on online ontologies (Semantic Web) to derive mappings </li></ul><ul><li>ontologies are dynamically discovered (using Watson) and combined </li></ul>A B rel Semantic Web Does not rely on any pre-selected knowledge sources. Sabou, M., d'Aquin, M., and Motta, E. (2008) Exploring the Semantic Web as Background Knowledge for Ontology Matching , Journal of Data Semantic, XI.
    57. 57. The Question is … How to combine online ontologies to derive mappings?
    58. 58. Strategy 1 - Definition Find ontologies that contain equivalent classes for A and B and use their relationship in the ontologies to derive the mapping. A B rel Semantic Web A 1 ’ B 1 ’ A 2 ’ B 2 ’ A n ’ B n ’ O 1 O 2 O n For each ontology use these rules: … These rules can be extended to take into account indirect relations between A’ and B’, e.g., between parents of A’ and B’:
    59. 59. Strategy 1- Examples But what if there exists no ontology that contains both A and B? ka2.rdf Researcher AcademicStaff Semantic Web Researcher AcademicStaff ISWC SWRC Beef Food Semantic Web Beef RedMeat Tap Food MeatOrPoultry SR-16 FAO_Agrovoc
    60. 60. Strategy 2 - Definition Principle: If no ontologies are found that contain the two terms then combine information from multiple ontologies to find a mapping. A B rel Semantic Web A’ B C C’ B’ rel rel Details: (1) Select all ontologies containing A’ equiv. with A (2) For each ontology containing A’: (a) if find relation between C and B. (b) if find relation between C and B. Details: (1) Select all ontologies containing A’ equiv. with A (2) For each ontology containing A’: (a) if find relation between C and B. (b) if find relation between C and B.
    61. 61. Strategy 2 - Examples Vs. (midlevel-onto) (Tap) Ex1: Vs. Ex2: (r1) (pizza-to-go) (SUMO) (Same results for Duck, Goose, Turkey) (r1) Vs. Ex3: (pizza-to-go) (wine.owl) (r3)
    62. 62. Large Scale Evaluation <ul><li>Ontology alignment evaluation initiative - food </li></ul><ul><li>AGROVOC: UN FAO’s agricultural thesaurus </li></ul><ul><ul><li>±16.000 terms </li></ul></ul><ul><ul><li>multilingual </li></ul></ul><ul><li>NALT: United States National Agricultural Library Thesaurus </li></ul><ul><ul><li>±41.000 terms </li></ul></ul><ul><li>Precision obtained: 70% </li></ul>
    63. 63. <ul><li>SCARLET - relation discovery on the SW </li></ul><ul><li>http://scarlet.open.ac.uk/ </li></ul><ul><li>Automatically selects and combines multiple online ontologies to derive a relation </li></ul>Basic functionality used: Relation Discovery Concept_A (e.g., Supermarket) Concept_B (e.g., Building) Scarlet Semantic Web Semantic Relation ( ) Deduce Access
    64. 65. Watson <ul><li>Infrastructures for storing large scale semantics (Watson) </li></ul><ul><li>The story here is that Watson could be used as a starting point for building a storage infrastructure for the distributed semantic information in SmartProducts. </li></ul>
    65. 66. is a Search Engine for the Semantic Web Gateway
    66. 67. Architecture
    67. 68. Web Interface
    68. 69. Web Interface Advanced Keyword Search
    69. 70. Web Interface Ontology Exploration
    70. 71. Web Interface Ontology Metadata
    71. 72. Web Interface Querying
    72. 73. APIs <ul><li>SOAP and REST APIs that provide the infrastructure to: </li></ul><ul><ul><li>F ind SW documents and retrieve metadata about them </li></ul></ul><ul><ul><li>Find entities (classes, properties, individuals) and explore their semantic description </li></ul></ul><ul><ul><li>Apply SPARQL queries to Semantic Web documents </li></ul></ul>
    73. 74. Next Generation Semantic Web Applications WATSON enables a new generation of Semantic Web applications that need to access and reuse semantic information distributed on the entire Web.
    74. 75. Examples of NGSW
    75. 76. IEEE Intelligent Systems 23(3), pp. 20-28, May/June 2008 <ul><li>Key aspects of the paradigm </li></ul><ul><li>Tech. Infrastructure </li></ul><ul><li>Concrete Applications </li></ul>
    76. 77. PoweAqua <ul><li>Tools for exploiting distributed semantics </li></ul><ul><ul><li>Open Domain, Multi-Ontology Question Answering (PoweAqua) </li></ul></ul>
    77. 78. PowerAqua <ul><li>PowerAqua </li></ul><ul><ul><li>Cross ontology question aswering </li></ul></ul><ul><ul><li>Selects and combines relevant </li></ul></ul><ul><ul><li>information from multiple ontologie </li></ul></ul>
    78. 79. PowerAqua Open domain QA by exploring distributed semantic data. Natural language question Answers from online semantic data
    79. 80. PowerAqua: Architecture <ul><li>Steps 2 and 3 implement a run time knowledge matcher that efficiently produces mappings across ontologies and domains </li></ul><ul><ul><ul><li>Performs concepts, relations, instances and literal mapping </li></ul></ul></ul><ul><ul><ul><li>No assumptions on the user input </li></ul></ul></ul><ul><ul><ul><li>No assumptions on the domain, structure or complexity of ontologies </li></ul></ul></ul>