NIFify: Towards Better Quality Entity Linking Datasets Henry - - PowerPoint PPT Presentation

nifify towards better quality entity linking
SMART_READER_LITE
LIVE PREVIEW

NIFify: Towards Better Quality Entity Linking Datasets Henry - - PowerPoint PPT Presentation

NIFify: Towards Better Quality Entity Linking Datasets Henry Rosales-M endez, Aidan Hogan and Barbara Poblete University of Chile { hrosales,ahogan,bpoblete } @dcc.uchile.cl May 14, 2019 LA-WEB 2019 - 10th Latin American Web Congress


slide-1
SLIDE 1

NIFify: Towards Better Quality Entity Linking Datasets †

Henry Rosales-M´ endez, Aidan Hogan and Barbara Poblete

University of Chile {hrosales,ahogan,bpoblete}@dcc.uchile.cl

May 14, 2019

† LA-WEB 2019 - 10th Latin American Web Congress

slide-2
SLIDE 2

Example

slide-3
SLIDE 3

Example - Entity Recognition

slide-4
SLIDE 4

Example - Entity Disambiguation

slide-5
SLIDE 5

Name Variations in Entity Linking

Michael J. Jackson King of Pop Michael Joseph Jackson

slide-6
SLIDE 6

Name Variations in Entity Linking

Michael Jackson

slide-7
SLIDE 7

Are there benchmark datasets to measure Entity Linking results?

slide-8
SLIDE 8

Overview of popular EL datasets

Dataset Mn Typ Format MSNBC ✗ ✗ MSNBC IITB ✓ ✗ IITB AIDA/CoNLL ✓ ✗ AIDA ACE2004 ✗ ✗ MSNBC AQUAINT ✗ ✗ MSNBC DBpedia Spotlight ✓ ✗ Lexvo KORE50 ✓ ✗ AIDA N3-RSS 500 ✓ ✗ NIF Reuters 128 ✓ ✗ NIF News-100 ✓ ✗ NIF Wes2015 ✓ ✗ NIF SemEval 2015 Task 13 ✓ ✗ SemEval Thibaudet ✗ ✓ RENDEN Bergson ✗ ✓ RENDEN DBpedia Abstracts ✗ ✗ NIF MEANTIME ✓ ✓ CAT VoxEL ✓ ✗ NIF

slide-9
SLIDE 9

Overview of popular EL datasets

Dataset Mn Typ Format MSNBC ✗ ✗ MSNBC IITB ✓ ✗ IITB AIDA/CoNLL ✓ ✗ AIDA ACE2004 ✗ ✗ MSNBC AQUAINT ✗ ✗ MSNBC DBpedia Spotlight ✓ ✗ NIF KORE50 ✓ ✗ NIF N3-RSS 500 ✓ ✗ NIF Reuters 128 ✓ ✗ NIF News-100 ✓ ✗ NIF Wes2015 ✓ ✗ NIF SemEval 2015 Task 13 ✓ ✗ SemEval Thibaudet ✗ ✓ RENDEN Bergson ✗ ✓ RENDEN DBpedia Abstracts ✗ ✗ NIF MEANTIME ✓ ✓ CAT VoxEL ✓ ✗ NIF

slide-10
SLIDE 10

Proposal

NIFify: a tool that simultaneously supports the creation, visualization, and validation of NIF datasets, as well as the comparison of EL systems.

slide-11
SLIDE 11

Related Work

  • NIF-Dataset creation

BENGAL Automatic NIF Creation

Demo Source Code QRTool

slide-12
SLIDE 12

Related Work

  • NIF-Dataset creation

BENGAL Automatic NIF Creation

Demo Source Code QRTool

  • NIF-Dataset validation

Eaglet NIF-Dataset Validation Demo Source Code

slide-13
SLIDE 13

Related Work

  • NIF-Dataset creation

BENGAL Automatic NIF Creation

Demo Source Code QRTool

  • NIF-Dataset validation

Eaglet NIF-Dataset Validation Demo Source Code

  • Benchmarking
NIF-Dataset Creation Orbis Benchmark Demo Source Code Visualization GERBIL Benchmark Demo Source Code Visualization
slide-14
SLIDE 14

Related Work

  • NIF-Dataset creation

BENGAL Automatic NIF Creation

Demo Source Code QRTool

  • NIF-Dataset validation

Eaglet NIF-Dataset Validation Demo Source Code

  • Benchmarking
NIF-Dataset Creation Orbis Benchmark Demo Source Code Visualization GERBIL Benchmark Demo Source Code Visualization

NIF-Dat

slide-15
SLIDE 15

NIFify - Creation

slide-16
SLIDE 16

NIFify - Creation

slide-17
SLIDE 17

NIFify - Creation

slide-18
SLIDE 18

NIFify - Creation

slide-19
SLIDE 19

NIFify - Validation

slide-20
SLIDE 20

NIFify - Validation

slide-21
SLIDE 21

Errors found in current NIF datasets

Dataset Spelling Error Link Error Format Error DBpedia Spotlight 8 23 4 N3-RSS 500 1 34 – Reuters 128 4 71 – News-100 9 1515 – Wes2015 – 609 – VoxEL – 8 –

  • https://users.dcc.uchile.cl/~hrosales/dataset_errors.html
slide-22
SLIDE 22

Errors found in current NIF datasets

slide-23
SLIDE 23

NIFify - Benchmarking

slide-24
SLIDE 24

NIFify - Benchmarking

slide-25
SLIDE 25

Conclusion

  • NIFify: Creation/Validation/Visualization/Benchmark
  • Demo:

https://users.dcc.uchile.cl/~hrosales/NIFify_v2.html

  • Source Code:

https://github.com/henryrosalesmendez/NIFify_v2

slide-26
SLIDE 26

NIFify: Towards Better Quality Entity Linking Datasets †

Henry Rosales-M´ endez, Aidan Hogan and Barbara Poblete

University of Chile {hrosales,ahogan,bpoblete}@dcc.uchile.cl

May 14, 2019

† LA-WEB 2019 - 10th Latin American Web Congress