Automatic Itinerary Reconstruction from Texts L. Moncla (LIUPPA, - - PowerPoint PPT Presentation

▶
automatic itinerary reconstruction from texts
SMART_READER_LITE
LIVE PREVIEW

Automatic Itinerary Reconstruction from Texts L. Moncla (LIUPPA, - - PowerPoint PPT Presentation

Feb 26, 2023 300 likes •875 views

Automatic Itinerary Reconstruction from Texts L. Moncla (LIUPPA, IAAA), M. Gaio (LIUPPA), S. Mustire (COGIT) L. Moncla ludovic.moncla@univ-pau.fr GIScience 2014 GIScience 2014 2/56 Automatic Itinerary Reconstruction from Texts L.

slide-1
SLIDE 1
  • L. Moncla

ludovic.moncla@univ-pau.fr

GIScience 2014

Automatic Itinerary Reconstruction from Texts

  • L. Moncla (LIUPPA, IAAA), M. Gaio (LIUPPA), S. Mustière (COGIT)
slide-2
SLIDE 2

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 2/56

CONTENTS

1

Introduction

2

Solution adopted

3

Implementation

4

Conclusion

slide-3
SLIDE 3

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 3/56

OUTLINE

1

Introduction Objectives Corpus

2

Solution adopted

3

Implementation

4

Conclusion

slide-4
SLIDE 4

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 4/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Objectives

Traverser Champagny-le-Haut et contourner le hameau de Friburge. Vous apercevrez le Lac de la Plagne puis marcher jusqu’au refuge au sud du lac de Grattaleu.

slide-5
SLIDE 5

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 4/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Objectives

Traverser Champagny-le-Haut et contourner le hameau de Friburge. Vous apercevrez le Lac de la Plagne puis marcher jusqu’au refuge au sud du lac de Grattaleu.

Who understand ?

slide-6
SLIDE 6

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 4/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Objectives

Cross Champagny-le-Haut and get around from the north of hamlet

  • Friburge. You will see the Lac de la Plagne then walk to the refuge

south of lake Grattaleu.

better ? Where is it ?

slide-7
SLIDE 7

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 4/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Objectives

and now ?

slide-8
SLIDE 8

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 5/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Objectives

From text... Traverser Champagny-le-Haut et contourner le hameau de Friburge. Vous apercevrez le Lac de la Plagne puis marcher jusqu’au refuge au sud du lac de Grattaleu. ...to map.

slide-9
SLIDE 9

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 6/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Objectives

Automatic itinerary reconstruction from texts

Main problems

1 Expression of space and motion in language 2 Automatic information extraction 3 Assigning location to spatial information 4 Route calculation

slide-10
SLIDE 10

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 7/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Objectives

Main problems

1 Expression of space and motion in language [Talmy,1985]

  • spatial named entities (toponyms)
  • geographical terms
  • spatial relation
  • spatial prepositions, spatial adverb, etc
  • verbs of visual perception
  • verbs of displacement, motion or location
slide-11
SLIDE 11

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 8/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Objectives

Main problems

2 Automatic information extraction

  • Named Entity Recognition (NER)
  • learning techniques, ad-hoc rules, etc
  • hybrid systems based on syntactico-semantic patterns
  • types of named entities : date, person, spatial, etc
  • Natural Language Processing (NLP)
  • Part-of-speech (POS) tagging
slide-12
SLIDE 12

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 9/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Objectives

Main problems

3 Assigning location to spatial information

  • Toponym resolution (Geocoding)
  • use of geographical resources
  • Toponym disambiguation [Smith and Mann, 2003]
  • Referent ambiguity : the same name used for several places
  • Reference ambiguity : several names for the same place
  • Referent class ambiguity : geo / non-geo ambiguity
  • Unreferenced toponyms ambiguity
slide-13
SLIDE 13

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 10/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Objectives

Main problems

4 Route calculation

  • Route planning
  • Spatial analysis
  • using topological, geometric, or geographic properties
  • Spatio-temporal analysis
slide-14
SLIDE 14

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 11/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Corpus

Corpus of experiments

  • 1295 French hike descriptions
  • automatically collected from the Web
  • collected with GPS track

Body of reference

  • 30 French hike descriptions manually annotated
  • controlled manual tagging tool under development
slide-15
SLIDE 15

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 11/40

Contents Introduction

Objectives Corpus

Solution adopted Implementation Conclusion

Corpus

Corpus of experiments

  • 1295 French hike descriptions
  • automatically collected from the Web
  • collected with GPS track

Body of reference

  • 30 French hike descriptions manually annotated
  • controlled manual tagging tool under development
slide-16
SLIDE 16

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 12/56

OUTLINE

1

Introduction

2

Solution adopted Geoparsing Itinerary calculation

3

Implementation

4

Conclusion

slide-17
SLIDE 17

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 13/40

Contents Introduction Solution adopted

Geoparsing Itinerary calculation

Implementation Conclusion

Geoparsing

Automatic annotation of spatial expressions

Expanded spatial named entities (ESNE)

  • Spatial named entities [mandatory]
  • Toponym + sub-type (feature types)
  • e.g. Vienna, Danube river, lake Grattaleu, etc
  • Geographical terms [optional]
  • e.g. refuge, bridge, hamlet, etc
  • Indirections [optional]
  • e.g. the north of, from the left, etc

1 : left side of the Danube river 2 : the refuge south of lake Grattaleu

slide-18
SLIDE 18

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 13/40

Contents Introduction Solution adopted

Geoparsing Itinerary calculation

Implementation Conclusion

Geoparsing

Automatic annotation of spatial expressions

Expanded spatial named entities (ESNE)

  • Spatial named entities [mandatory]
  • Toponym + sub-type (feature types)
  • e.g. Vienna, Danube river, lake Grattaleu, etc
  • Geographical terms [optional]
  • e.g. refuge, bridge, hamlet, etc
  • Indirections [optional]
  • e.g. the north of, from the left, etc

1 : left side of the Danube river 2 : the refuge south of lake Grattaleu

slide-19
SLIDE 19

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 14/40

Contents Introduction Solution adopted

Geoparsing Itinerary calculation

Implementation Conclusion

Geoparsing

Automatic annotation of spatial expressions

Expression of motion

  • motion verbs
  • to go, to leave, etc
  • verbs of visual perception
  • to see, to glimpse, etc
  • "topographic" verbs
  • to converge, to overhang, etc
  • location verbs
  • to locate, to be, etc

VTo structures

  • relationship between toponyms or ESNE and expressions of motion

1 : Reach the left side of the Danube river 2 : Walk to the refuge south of lake Grattaleu Polarity of motion verbs Initial verbs : Final verbs : Median verbs :

slide-20
SLIDE 20

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 14/40

Contents Introduction Solution adopted

Geoparsing Itinerary calculation

Implementation Conclusion

Geoparsing

Automatic annotation of spatial expressions

Expression of motion

  • motion verbs
  • to go, to leave, etc
  • verbs of visual perception
  • to see, to glimpse, etc
  • "topographic" verbs
  • to converge, to overhang, etc
  • location verbs
  • to locate, to be, etc

VTo structures

  • relationship between toponyms or ESNE and expressions of motion

1 : Reach the left side of the Danube river 2 : Walk to the refuge south of lake Grattaleu

slide-21
SLIDE 21

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 14/40

Contents Introduction Solution adopted

Geoparsing Itinerary calculation

Implementation Conclusion

Geoparsing

Automatic annotation of spatial expressions

Expression of motion

  • motion verbs
  • to go, to leave, etc
  • verbs of visual perception
  • to see, to glimpse, etc
  • "topographic" verbs
  • to converge, to overhang, etc
  • location verbs
  • to locate, to be, etc

VTo structures

  • relationship between toponyms or ESNE and expressions of motion

1 : Reach the left side of the Danube river 2 : Walk to the refuge south of lake Grattaleu

slide-22
SLIDE 22

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 14/40

Contents Introduction Solution adopted

Geoparsing Itinerary calculation

Implementation Conclusion

Geoparsing

Automatic annotation of spatial expressions

Expression of motion

  • motion verbs
  • to go, to leave, etc
  • verbs of visual perception
  • to see, to glimpse, etc
  • "topographic" verbs
  • to converge, to overhang, etc
  • location verbs
  • to locate, to be, etc

VTo structures

  • relationship between toponyms or ESNE and expressions of motion

1 : Reach the left side of the Danube river 2 : Walk to the refuge south of lake Grattaleu

slide-23
SLIDE 23

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 14/40

Contents Introduction Solution adopted

Geoparsing Itinerary calculation

Implementation Conclusion

Geoparsing

Automatic annotation of spatial expressions

Expression of motion

  • motion verbs
  • to go, to leave, etc
  • verbs of visual perception
  • to see, to glimpse, etc
  • "topographic" verbs
  • to converge, to overhang, etc
  • location verbs
  • to locate, to be, etc

VTo structures

  • relationship between toponyms or ESNE and expressions of motion

1 : Reach the left side of the Danube river 2 : Walk to the refuge south of lake Grattaleu

slide-24
SLIDE 24

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 15/40

Contents Introduction Solution adopted

Geoparsing Itinerary calculation

Implementation Conclusion

Itinerary calculation

Combine language and spatial analysis to order places and reconstruct the path

  • Use information extracted from textual description
  • places not reached (perception, negation,...)
  • indirection (direction, distance,...)
  • motion expressions, polarity of the displacement (find origin and

destination)

  • Use spatial analysis to link places
slide-25
SLIDE 25

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 16/40

Contents Introduction Solution adopted

Geoparsing Itinerary calculation

Implementation Conclusion

Itinerary calculation

Minimum weight spanning tree

  • Geoparsing
  • polarity of displacement
  • places not reached
  • Geocoding
  • distances

FIGURE : Minimum weight spanning tree (source : wikipedia)

slide-26
SLIDE 26

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 17/56

OUTLINE

1

Introduction

2

Solution adopted

3

Implementation Geoparsing Itinerary calculation

4

Conclusion

slide-27
SLIDE 27

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 18/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Implementation

FIGURE : Block diagram of our processing chain

slide-28
SLIDE 28

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 19/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Geoparsing

Finite-state transducers cascade

Our linguistic rules are implemented using a finite-state transducers cascade Input

  • Raw text + POS tags

example : {Walk,.V} {to,.PREP} {the,.ART} {refuge,.N} {south,.N} {of,.PREP} {lake,.N} {Grattaleu,.NPr} {.,.PUN} Main transducers

1 Indirections (spatial relations) 2 Candidate toponyms (+ sub-type) 3 ESNE 4 Motion, perception (VTo Structures)

slide-29
SLIDE 29

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 19/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Geoparsing

Finite-state transducers cascade

Our linguistic rules are implemented using a finite-state transducers cascade Input

  • Raw text + POS tags

example : {Walk,.V} {to,.PREP} {the,.ART} {refuge,.N} {south,.N} {of,.PREP} {lake,.N} {Grattaleu,.NPr} {.,.PUN} Main transducers

1 Indirections (spatial relations) 2 Candidate toponyms (+ sub-type) 3 ESNE 4 Motion, perception (VTo Structures)

slide-30
SLIDE 30

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 20/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Geoparsing

Finite-state transducers cascade

Example of transducer

FIGURE : Example of transducer in the Unitex platform

slide-31
SLIDE 31

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 21/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Result of geoparsing

Cross Champagny-le-Haut and get around from the north of hamlet

  • Friburge. You will see the Lac de la Plagne then walk to the refuge

south of lake Grattaleu. Execution of the transducers cascade

1 Indirections (spatial relations) 2 Candidate toponyms (+ sub-type) 3 ESNE 4 Motion, perception (VTo Structures)

slide-32
SLIDE 32

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 21/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Result of geoparsing

Cross Champagny-le-Haut and get around from the north of hamlet

  • Friburge. You will see the Lac de la Plagne then walk to the refuge

south of lake Grattaleu. Execution of the transducers cascade

1 Indirections (spatial relations) 2 Candidate toponyms (+ sub-type) 3 ESNE 4 Motion, perception (VTo Structures)

slide-33
SLIDE 33

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 21/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Result of geoparsing

Cross Champagny-le-Haut and get around from the north of hamlet

  • Friburge. You will see the Lac de la Plagne then walk to the refuge

south of lake Grattaleu. Execution of the transducers cascade

1 Indirections (spatial relations) 2 Candidate toponyms (+ sub-type) 3 ESNE 4 Motion, perception (VTo Structures)

slide-34
SLIDE 34

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 21/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Result of geoparsing

Cross Champagny-le-Haut and get around from the north of hamlet

  • Friburge. You will see the Lac de la Plagne then walk to the refuge

south of lake Grattaleu. Execution of the transducers cascade

1 Indirections (spatial relations) 2 Candidate toponyms (+ sub-type) 3 ESNE 4 Motion, perception (VTo Structures)

slide-35
SLIDE 35

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 21/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Result of geoparsing

Cross Champagny-le-Haut and get around from the north of hamlet

  • Friburge. You will see the Lac de la Plagne then walk to the refuge

south of lake Grattaleu. Execution of the transducers cascade

1 Indirections (spatial relations) 2 Candidate toponyms (+ sub-type) 3 ESNE 4 Motion, perception (VTo Structures)

slide-36
SLIDE 36

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 22/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Result of geoparsing

Cross Champagny-le-Haut and get around from the north of hamlet

  • Friburge. You will see the Lac de la Plagne then walk to the refuge

south of lake Grattaleu.

  • Cross Champagny-le-Haut
  • get around from the north of hamlet Friburge
  • see the Lac de la Plagne
  • walk to the refuge south of lake Grattaleu
slide-37
SLIDE 37

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 23/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Geoparsing

Finite-state transducers cascade

Preprocessing (a) (b) (c) Precision Recall Automatic POS 583 581 595 90.02% 94.68% POS manually corrected 583 559 636 98.29% 99.42%

TABLE : Evaluation of our geoparsing process

(a) number of toponyms manually annotated (b) number of relevant toponyms automatically annotated (c) number of toponyms automatically annotated

slide-38
SLIDE 38

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 24/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Itinerary calculation

Geocoding

Querying of gazetteers

  • BDNyme : French toponymic database provided by IGN
  • Geonames : geographical database (geocoding Web Services)

Toponyms ambiguities

  • 50% of toponyms found in gazetteers have more than 1 result
  • 30% of toponyms are not found in gazetteers (problem of

fine-grain toponyms)

slide-39
SLIDE 39

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 25/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Itinerary calculation

Geocoding

Referent ambiguity : same name used for several places

slide-40
SLIDE 40

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 26/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Itinerary calculation

Geocoding

Structural ambiguity : ambiguity on words constituting the name

  • 46% of toponyms are associated with feature types

FR EN Frequency col pass 20 village town 20 hameau hamlet 20 rue road 17 chemin path 15 chalet cottage 13 refuge refuge 11 pont bridge 11 lac lake 8 chapelle chapel 8

TABLE : Most frequent terms associated with toponyms

slide-41
SLIDE 41

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 27/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Itinerary calculation

Geocoding

Toponyms desambiguation Result of geocoding + real GPS track

slide-42
SLIDE 42

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 28/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Itinerary calculation

Automatic itinerary reconstruction

Minimum spanning tree (euclidian distances)

slide-43
SLIDE 43

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 29/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Itinerary calculation

Automatic itinerary reconstruction

Minimum spanning tree (euclidian distances)

slide-44
SLIDE 44

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 30/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Itinerary calculation

Automatic itinerary reconstruction

Minimum spanning tree using information extracted from the text

slide-45
SLIDE 45

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 31/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Itinerary calculation

Automatic itinerary reconstruction

Minimum spanning tree using information extracted from the text

slide-46
SLIDE 46

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 32/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Itinerary calculation

Automatic itinerary reconstruction

Minimum spanning tree using information extracted from the text

slide-47
SLIDE 47

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 33/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Itinerary calculation

Automatic itinerary reconstruction

Minimum spanning tree using information extracted from the text

FIGURE : Automatic itinerary VS real GPS trace

slide-48
SLIDE 48

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 34/40

Contents Introduction Solution adopted Implementation

Processing chain Geoparsing Itinerary calculation Online demonstration

Conclusion

Online geoparsing and geocoding tool

http ://erig.univ-pau.fr/PERDIDO/

FIGURE : Screenshot of the online geoparsing and geocoding tool

slide-49
SLIDE 49

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 35/56

OUTLINE

1

Introduction

2

Solution adopted

3

Implementation

4

Conclusion

slide-50
SLIDE 50

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 36/40

Contents Introduction Solution adopted Implementation Conclusion

Conclusion

Expanded geoparsing process

  • spatial context of toponyms is extracted : feature types,

displacement, perception, etc Geocoding and itinerary reconstruction

  • minimum spanning tree weighted using information (places not

reached, etc)

  • information extracted from the text allows a better interpretation

Mixing spatial and textual analysis

  • first results are encouraging
slide-51
SLIDE 51

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 37/40

Contents Introduction Solution adopted Implementation Conclusion

Work in progress

Multilingual

  • French, Spanish, Italian

Toponyms desambiguation

  • Clustering algorithm : DBSCAN (Density-Based Spatial Clustering of

Applications with Noise)

  • Geocoding of unreferenced toponyms

Full paper accepted in ACM SIGSPATIAL 2014

slide-52
SLIDE 52

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 37/40

Contents Introduction Solution adopted Implementation Conclusion

Work in progress

Multilingual

  • French, Spanish, Italian

Toponyms desambiguation

  • Clustering algorithm : DBSCAN (Density-Based Spatial Clustering of

Applications with Noise)

  • Geocoding of unreferenced toponyms

Full paper accepted in ACM SIGSPATIAL 2014

slide-53
SLIDE 53

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 38/40

Contents Introduction Solution adopted Implementation Conclusion

Outlook

Improve the route calculation using other information

  • level of the trail
  • means of transport (walk, bike, etc)
  • similarity, complementarity and proximity
  • temporal information
slide-54
SLIDE 54

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 39/40

Contents Introduction Solution adopted Implementation Conclusion

Outlook

The use of temporal information

Preliminary experiments

  • Ordering toponyms as thez occur in the text
  • The need of temporal context
slide-55
SLIDE 55

Automatic Itinerary Reconstruction from Texts

  • L. Moncla, M. Gaio, and S. Mustière

GIScience 2014 – 39/40

Contents Introduction Solution adopted Implementation Conclusion

Outlook

The use of temporal information

Preliminary experiments

  • Ordering toponyms as thez occur in the text
  • The need of temporal context
slide-56
SLIDE 56

Thank you for your attention

CONTACT

  • L. Moncla

ludovic.moncla@univ-pau.fr http ://erig.univ-pau.fr/PERDIDO/