Camille Tardy, PhD Defence
January, 30th
2017
University of Geneva
INTRODUCING SPATIAL
COVERAGE IN A SEMANTIC
REPOSITORY M...
‣ Introduction
‣ Research Questions
‣ State of the Art
‣ Repository Model
‣ Associating Documents and City Objects
‣ Extra...
INTRODUCTION
3
4
http://www.worldsmartcity.org
INTRODUCTION
➤ Digital Libraries (DL) : handles resources
➤ Geographic Information System (GIS) : handles map
➤ Connect bo...
INTRODUCTION
➤ Documents validity depends on a given time and space
➤ Scope of ressources
Introduction 6
RESEARCH
QUESTIONS
7
“How to present, qualify and define
the geo-spatial context of any
resource: texts, images, datasets, 3D
models etc. in dig...
“How to validate this model on
complex use cases?
How to use this model to localise
documents in 2D or 3D city maps,
and t...
STATE OF THE
ART
10
STATE OF THE ART
➤ Two aspects:
➤ Spatial management of documents
➤ Spatial visualisation of documents
State of the Art 11
GIS AND GIR
➤ GIS and Geographic Information Retrieval Systems (GIR)
➤ Geospatial and temporal contextualisation of inform...
EXAMPLES
➤ GeoTracker [2]
➤ Each document
related to N places
State of the Art 13
EXAMPLES
➤ PIV [3]
➤ Document index as a polygon
State of the Art 14
EXAMPLES
➤ World Explorer [4]
➤ Map enrichment
State of the Art 15
EXAMPLES
➤ GIPSY [5]
➤ 3D Indexing
State of the Art 16
INDEXING AND MATCHING
➤ Select/identify geo entities in resource
➤ Index resource
Geo identification : difficult task in N...
VISUALISATION OF DOCUMENTS
➤ Spatial visualisation of documents
➤ CityGML links between geometry and semantic
State of the...
REPOSITORY
MODEL
19
GLOBAL VIEW
Repository Model 20
ANNOTATION VOCABULARY MODEL
21Repository Model
Ecology
Transportation
Architecture
ANNOTATION VOCABULARY MODEL
22Repository Model
DOCUMENT MODEL
Repository Model 23
SPATIAL RESSOURCE MODEL
➤ The spatial resource model is composed of :
➤ Object description
➤ Semantic / Identification —> a...
COVERAGE
25
“
COVERAGE DEFINITION
The geographic and temporal context
of a resource refers to the spatial and
temporal regions in whic...
COVERAGE MODEL
Repository Model 27
Hospital
HUG
COVERAGE DEFINITION
➤ Composition of entities : place names, functions or properties
➤ Use semantic vocabularies
➤ Class d...
COVERAGE DEFINITION
➤ A resource that has sub-resources:
coverage of the parent ≥ { ∪ coverage of the sub-resources }
➤ No...
COVERAGE DEFINITION FOR DIFFERENT RESOURCE TYPE
Repository Model 30
historical
event
photo
Specific coverage
Universal cov...
PHOTO-EVENT COVERAGE EXAMPLE
31Repository Model
Coverage : Hotel du parc (March 18th, 1962 - Today)
Content : Algeria ∪ Fr...
QUERY MODEL
Repository Model 32
Q1(Content: “brain surgery”;
Coverage: Geneva)
➤ Filter the documents that have a spatial coverage that matches
the query’s coverage
➤ Coverage feature (CF), coverage cl...
GEOGRAPHIC COVERAGE IN WEB SEARCH QUERIES
➤ Query: “legislation eau potable” (drinking water laws)
➤ Coverage: “Switzerlan...
GEOGRAPHIC COVERAGE IN WEB SEARCH QUERIES
35
➤ Evaluate result ranking by expert
• Discounted Cumulated Gain (DCG) [6]
ASSOCIATING
DOCUMENTS AND
CITY OBJECTS
36
TO BUILD THE REPOSITORY MODEL
➤ Annotation Vocabulary
➤ Time vocabulary
➤ Spatial vocabulary
➤ Bridge Domain ontology
➤ De...
IMPLEMENTATION
38
BUILDING THE ANNOTATION MODEL
➤ Core of the annotation ontology
• Space : Geonames ontology (geographic aspect) and
CityGM...
ANNOTATING THE SPATIAL RESOURCE
➤ Identification of city objects based on footprints and semantic
Associating Documents and...
LOCATE RESOURCES
➤ Align documents and spatial objects
Associating Documents and City Objects 41
MATRIX : DOCUMENT-OBJECT LINK
➤ Geo coverage comparison
➤ Generates a matrix : M(doc,obj) [0,1]
1. A direct link
• Same en...
MATRIX : DOCUMENT-OBJECT LINK
3. Link through non-geographic entities
• Domain ontology
• Semantic distance
4. Link by spa...
EXTRACTING
GEOGRAPHIC
INFORMATION
FROM TAGS
44
METHOD PRESENTATION
➤ Non-statistical
➤ Knowledge based approach using existing Volunteered Geographic
Information (VGI) s...
METHOD PRESENTATION
➤ Distinction between the spatial coverage and the spatial
content of a document
➤ Photo coverage : “w...
CATEGORISATION EXAMPLE
47Extracting Geographic Information from Tags
Unidentified
Colour
Geographic feature / feature clas...
ALGORITHM
48Extracting Geographic Information from Tags
Tag ti
geo
process
word sense
process
Geo weight
gw(ti)
Sense & Ca...
DISAMBIGUATION EXAMPLE
49Extracting Geographic Information from Tags
PROTOTYPE VIEW
Extracting Geographic Information from Tags 50
VALIDATION
➤ For all the photos:
• Precision = 72.5%
• Recall = 66.7%
• F-measure = 0.695 

➤ Error sources:
➤ Crowdsourci...
CONCLUSION
AND
FUTURE WORKS
52
CONCLUSION
➤ Model of Digital Library
➤ Coverage model
➤ Coverage centred, extensible semantic annotation model
➤ Validati...
FUTURE WORKS
54
FUTURE WORKS
➤ Precisely evaluate first implementation of the model
➤ precision recall for document-object association
➤ Ex...
REFERENCES & ANNEXES
56
REFERENCES
1. C. B. Jones and R. S. Purves, “Geographical information retrieval,”
International Journal of Geographical In...
REFERENCES
5. A. Woodruff and C. Plaunt, “GIPSY: Automated Geographic Indexing
of Text Documents,” JASIS, pp. 1–21, 1994.
6...
PERSONAL REFERENCES
a. N. Ghoula, H. de Ribaupierre, C. Tardy, and G. Falquet, “Opérations sur des ressources
hétérogènes ...
IMAGES CREDITS
✦ Titles
✦ Albert Robida, “Maison tournante aérienne”
✦ Matthew Hutchinson, “The Questioning Roboto” (https...
Prochain SlideShare
Chargement dans…5
×

Introducing Spatial Coverage in a Semantic Repository Model - Phd defence

85 vues

Publié le

University of Geneva - January 30th 2017
Introducing Spatial Coverage in a Semantic Repository Model
Camille Tardy

Publié dans : Sciences
0 commentaire
0 j’aime
Statistiques
Remarques
  • Soyez le premier à commenter

  • Soyez le premier à aimer ceci

Aucun téléchargement
Vues
Nombre de vues
85
Sur SlideShare
0
Issues des intégrations
0
Intégrations
2
Actions
Partages
0
Téléchargements
1
Commentaires
0
J’aime
0
Intégrations 0
Aucune incorporation

Aucune remarque pour cette diapositive

Introducing Spatial Coverage in a Semantic Repository Model - Phd defence

  1. 1. Camille Tardy, PhD Defence January, 30th 2017 University of Geneva INTRODUCING SPATIAL COVERAGE IN A SEMANTIC REPOSITORY MODEL 1
  2. 2. ‣ Introduction ‣ Research Questions ‣ State of the Art ‣ Repository Model ‣ Associating Documents and City Objects ‣ Extracting Geographic Information from Tags ‣ Conclusion and Future Work ‣ References 2
  3. 3. INTRODUCTION 3
  4. 4. 4 http://www.worldsmartcity.org
  5. 5. INTRODUCTION ➤ Digital Libraries (DL) : handles resources ➤ Geographic Information System (GIS) : handles map ➤ Connect both worlds : include spatial aspect in resources ➤ Our work focus on DL : Spatial dimension in DL Introduction 5
  6. 6. INTRODUCTION ➤ Documents validity depends on a given time and space ➤ Scope of ressources Introduction 6
  7. 7. RESEARCH QUESTIONS 7
  8. 8. “How to present, qualify and define the geo-spatial context of any resource: texts, images, datasets, 3D models etc. in digital libraries? 8Research Questions
  9. 9. “How to validate this model on complex use cases? How to use this model to localise documents in 2D or 3D city maps, and to semantically enhance geographic services? 9Research Questions
  10. 10. STATE OF THE ART 10
  11. 11. STATE OF THE ART ➤ Two aspects: ➤ Spatial management of documents ➤ Spatial visualisation of documents State of the Art 11
  12. 12. GIS AND GIR ➤ GIS and Geographic Information Retrieval Systems (GIR) ➤ Geospatial and temporal contextualisation of information ➤ Key issues from literature: [1] ➤ the detection disambiguation of geographic references ➤ spatial and textual indexing ➤ the interpretation of fuzzy geographic terminology ➤ geographic relevance ranking and interfaces State of the Art 12
  13. 13. EXAMPLES ➤ GeoTracker [2] ➤ Each document related to N places State of the Art 13
  14. 14. EXAMPLES ➤ PIV [3] ➤ Document index as a polygon State of the Art 14
  15. 15. EXAMPLES ➤ World Explorer [4] ➤ Map enrichment State of the Art 15
  16. 16. EXAMPLES ➤ GIPSY [5] ➤ 3D Indexing State of the Art 16
  17. 17. INDEXING AND MATCHING ➤ Select/identify geo entities in resource ➤ Index resource Geo identification : difficult task in Named Entity Recognition(NER) ➤ Query and resources footprint —> textual or geometric form ➤ Matching & Retrieval : based on footprints inclusions State of the Art 17
  18. 18. VISUALISATION OF DOCUMENTS ➤ Spatial visualisation of documents ➤ CityGML links between geometry and semantic State of the Art 18 pharmacy post office library office building road school park
  19. 19. REPOSITORY MODEL 19
  20. 20. GLOBAL VIEW Repository Model 20
  21. 21. ANNOTATION VOCABULARY MODEL 21Repository Model Ecology Transportation Architecture
  22. 22. ANNOTATION VOCABULARY MODEL 22Repository Model
  23. 23. DOCUMENT MODEL Repository Model 23
  24. 24. SPATIAL RESSOURCE MODEL ➤ The spatial resource model is composed of : ➤ Object description ➤ Semantic / Identification —> annotations ➤ The linking, querying and indexing are based on the annotations ➤ Description language + Geo-Spatial vocabulary ➤ e.g.: CityGML + Geonames Classes + Instances Repository Model 24
  25. 25. COVERAGE 25
  26. 26. “ COVERAGE DEFINITION The geographic and temporal context of a resource refers to the spatial and temporal regions in which the resource is true or must be true in the real world or in a fiction. 26Repository Model
  27. 27. COVERAGE MODEL Repository Model 27 Hospital HUG
  28. 28. COVERAGE DEFINITION ➤ Composition of entities : place names, functions or properties ➤ Use semantic vocabularies ➤ Class definition (logical expressions) as coverage Repository Model 28 coverage = {feature_code some Populated_place and population value 1000 and level1_Administrative_parent value Rhône-Alpes}
  29. 29. COVERAGE DEFINITION ➤ A resource that has sub-resources: coverage of the parent ≥ { ∪ coverage of the sub-resources } ➤ No coverage of sub-resource can be ≥ to the one of the whole resource Repository Model 29
  30. 30. COVERAGE DEFINITION FOR DIFFERENT RESOURCE TYPE Repository Model 30 historical event photo Specific coverage Universal coverage
  31. 31. PHOTO-EVENT COVERAGE EXAMPLE 31Repository Model Coverage : Hotel du parc (March 18th, 1962 - Today) Content : Algeria ∪ France (1954 - 1962). Signature of the Evian Agreements
  32. 32. QUERY MODEL Repository Model 32 Q1(Content: “brain surgery”; Coverage: Geneva)
  33. 33. ➤ Filter the documents that have a spatial coverage that matches the query’s coverage ➤ Coverage feature (CF), coverage classes (CC) and the content (Cont) M(Q,D) = a1 MCF(QCF,DCF) + a2 MCC(QCC,DCC) + a3 MCont(QCont,DCont) MATCHING Repository Model 33 Coverage matching
  34. 34. GEOGRAPHIC COVERAGE IN WEB SEARCH QUERIES ➤ Query: “legislation eau potable” (drinking water laws) ➤ Coverage: “Switzerland” ➤ MCF(QCF,DCF) • Hierarchy (QCF Switzerland- DCF Europe) —> MCF = 0.5 • Siblings (QCF Switzerland - DCF Quebec) —>MCF = 0.5÷5580 = 0.00009 Associating Documents and City Objects 34 Ranking Scale
  35. 35. GEOGRAPHIC COVERAGE IN WEB SEARCH QUERIES 35 ➤ Evaluate result ranking by expert • Discounted Cumulated Gain (DCG) [6]
  36. 36. ASSOCIATING DOCUMENTS AND CITY OBJECTS 36
  37. 37. TO BUILD THE REPOSITORY MODEL ➤ Annotation Vocabulary ➤ Time vocabulary ➤ Spatial vocabulary ➤ Bridge Domain ontology ➤ Define coverages for documents ➤ Annotated documents ➤ 2D or 3D spatial model ➤ Identified with Spatial vocabulary Repository Model 37
  38. 38. IMPLEMENTATION 38
  39. 39. BUILDING THE ANNOTATION MODEL ➤ Core of the annotation ontology • Space : Geonames ontology (geographic aspect) and CityGML ontology (spatial aspect) • Domain bridge : Urbamet thesaurus Associating Documents and City Objects 39 Ontology available in [e]
  40. 40. ANNOTATING THE SPATIAL RESOURCE ➤ Identification of city objects based on footprints and semantic Associating Documents and City Objects 40 CityGML Geonames
  41. 41. LOCATE RESOURCES ➤ Align documents and spatial objects Associating Documents and City Objects 41
  42. 42. MATRIX : DOCUMENT-OBJECT LINK ➤ Geo coverage comparison ➤ Generates a matrix : M(doc,obj) [0,1] 1. A direct link • Same entities 2. Direct link by class • Class hierarchy • Semantic distance Associating Documents and City Objects 42
  43. 43. MATRIX : DOCUMENT-OBJECT LINK 3. Link through non-geographic entities • Domain ontology • Semantic distance 4. Link by spatial proximity • Euclidean distance Associating Documents and City Objects 43
  44. 44. EXTRACTING GEOGRAPHIC INFORMATION FROM TAGS 44
  45. 45. METHOD PRESENTATION ➤ Non-statistical ➤ Knowledge based approach using existing Volunteered Geographic Information (VGI) sources ➤ Find the characteristics of geographic places ➤ Based on ➤ Spatial coverage ➤ Tags categorisation ➤ Semantic identification ➤ Complete the semantic gap on places and points of interest (POIs) Extracting Geographic Information from Tags 45
  46. 46. METHOD PRESENTATION ➤ Distinction between the spatial coverage and the spatial content of a document ➤ Photo coverage : “what spatial region we see in the resource” ➤ Process Flickr photo tags ➤ Identify each tag ➤ Categorised the tags [7,8] to discard them: ➤ Event, Temporal, Weather, Actors, Meta, Colour and Unidentified Extracting Geographic Information from Tags 46
  47. 47. CATEGORISATION EXAMPLE 47Extracting Geographic Information from Tags Unidentified Colour Geographic feature / feature class Temporal Event Actor
  48. 48. ALGORITHM 48Extracting Geographic Information from Tags Tag ti geo process word sense process Geo weight gw(ti) Sense & Category {ti, sensei, cati} Disambiguation : 1.Find nonGeo weight ngw(ti) 2.Compare weights Dispatch tag : Geo OR nonGeo Geo coverage Characteristics Discard
  49. 49. DISAMBIGUATION EXAMPLE 49Extracting Geographic Information from Tags
  50. 50. PROTOTYPE VIEW Extracting Geographic Information from Tags 50
  51. 51. VALIDATION ➤ For all the photos: • Precision = 72.5% • Recall = 66.7% • F-measure = 0.695 
 ➤ Error sources: ➤ Crowdsourcing : ➤ missing disambiguation ➤ false evaluation ➤ spelling of the tags Extracting Geographic Information from Tags 51 Source code available in [f] ➤ Subjective aspect ➤ Language
  52. 52. CONCLUSION AND FUTURE WORKS 52
  53. 53. CONCLUSION ➤ Model of Digital Library ➤ Coverage model ➤ Coverage centred, extensible semantic annotation model ➤ Validation of the model ➤ First implementation of the main components ➤ Prototype (extract geographic information from tags) ➤ User validation of the categorisation Conclusion and Future Works 53
  54. 54. FUTURE WORKS 54
  55. 55. FUTURE WORKS ➤ Precisely evaluate first implementation of the model ➤ precision recall for document-object association ➤ Extend the tag categorisation system ➤ Associate the found characteristics with the corresponding places ➤ Tune parameters ➤ Associate with statistical approach (pre-process) ➤ Implement fully the DL model in connection with an existing DL system ➤ Create an automated visualisation tool for documents in 3D city models Conclusion and Future Works 55
  56. 56. REFERENCES & ANNEXES 56
  57. 57. REFERENCES 1. C. B. Jones and R. S. Purves, “Geographical information retrieval,” International Journal of Geographical Information Science, vol. 22, no. 3, pp. 219–228, Mar. 2008. 2. Y.-F. R. Chen, G. Di Fabbrizio, D. Gibbon, S. Jora, B. Renger, and Bin Wei, “Geotracker: geospatial and temporal RSS navigation,” presented at the WWW '07: Proceedings of the 16th international conference on World Wide Web, 2007. 3. M. Gaio, C. Sallaberry, P. Etcheverry, C. Marquesuzaa, and J. Lesbegueries, “A global process to access documents’ contents from a geographical point of view,” sciencedirect.com, vol. 19, no. 1, pp. 3–23, Feb. 2008. 4. S. Ahern, M. Naaman, R. Nair, and J. Yang, “World explorer: visualizing aggregate data from unstructured text in geo-referenced collections,” Proceedings of the 7th ACM/IEEE-CS joint conference on Digital libraries, pp. 1– 10, 2007. References 57
  58. 58. REFERENCES 5. A. Woodruff and C. Plaunt, “GIPSY: Automated Geographic Indexing of Text Documents,” JASIS, pp. 1–21, 1994. 6. R. B. Yates and B. R. Neto, Modern Information Retrieval: The Concepts and Technology behind Search. Addison-Wesley Professional, 2011. 7. R. Purves, A. Edwardes, and J. Wood, “Describing place through user generated content”, presented at the In Proceedings of the workshop on Metadata Mining for Image Understanding (MMIU 2008), Sheffield, 2011, vol. 16, no. 9. 8. R. S. Purves, A. Edwardes, and M. Sanderson, “Describing the where – improving image annotation and search through geography”, http://eprints.whiterose.ac.uk/4566/, 2008. 
 References 58
  59. 59. PERSONAL REFERENCES a. N. Ghoula, H. de Ribaupierre, C. Tardy, and G. Falquet, “Opérations sur des ressources hétérogènes dans un entrepôt de données à base d’ontologie”, presented at the 4e édition des journées francophones sur les ontologies (JFO), Montréal, Canada, 2011, pp. 203–216. b. C. Tardy, L. Moccozet, and G. Falquet, “Semantic alignment of documents with 3D city models”, presented at the Usage, Usability, and Utility of 3D City Models (3U3D), 2012. c. C. Tardy, L. Moccozet, and G. Falquet, “A Simple Tags Categorization Framework Using Spatial Coverage to Discover Geospatial Semantics”, presented at the Proceedings of the 25th International Conference Companion on World Wide Web (WWW), Montréal, Québec, Canada, 2016, pp. 657–660. d. C. Tardy, G. Falquet, and L. Moccozet, “Semantic enrichment of places with VGI sources”, presented at the GIR'16 the 10th Workshop on Geographic Information Retrieval, Burlingame, California, USA, 2016. e. https://gitlab.com/CamilleTrd/GeoAnnotation-ontology f. https://gitlab.com/CamilleTrd/GIEFT References 59
  60. 60. IMAGES CREDITS ✦ Titles ✦ Albert Robida, “Maison tournante aérienne” ✦ Matthew Hutchinson, “The Questioning Roboto” (https://flic.kr/p/7VhPft) ✦ Seiichi Kusunoki, “Bunch of Papers” (https://flic.kr/p/dVqT5t) ✦ Arto Alanenpää, “Lego Duplo” ✦ Ann Saunders and John Schofield (eds), Tudor London: a map and a view (2001) ✦ Examples : ✦ Algerian delegation arrivals to the hotel, for the signature of the “Accords d'Evian”, © AFP/STF 1962 (slide 31) ✦ “VioleTT Pi” Ludtz (https://flic.kr/p/m9ZBPB) (slide 46) ✦ “Eva Muñoz” Alejandro Pérez (https://flic.kr/p/66Lxf8) (slide 48) ✦ “Street #9” Alonso Ormeño (https://flic.kr/p/qnvRLa) (slide 49) Credits 60

×