Diese Präsentation wurde erfolgreich gemeldet.
Wir verwenden Ihre LinkedIn Profilangaben und Informationen zu Ihren Aktivitäten, um Anzeigen zu personalisieren und Ihnen relevantere Inhalte anzuzeigen. Sie können Ihre Anzeigeneinstellungen jederzeit ändern.
Building Event-Centric Knowledge
Graphs from News
Marco Rospocher, Marieke van Erp, Piek Vossen, Antske Fokkens, Itziar
Al...
Event-Centric Knowledge Graphs (ECKGs)
• An ECKG represents changes in the world
• Can capture long-term developments and ...
Why ECKGs?
• Policy makers and information specialists need to know
what’s happening/what’s happened to an entity
• Curren...
Processing Pipeline
Step 1: Information Extraction at the Document Level
Step 2: Cross-
document event
coreference
opinion...
NLP Pipeline
• State-of-the-art pipeline
• English, Spanish, Italian & Dutch
• ‘Black box’ setup and individual modules av...
NLP Pipeline output
Cross-Lingual Event Extraction
Cross-Lingual Event Extraction
Text2RDF
Event and Situation Ontology (ESO)
Event and Situation Ontology (ESO)
See also: P. Vossen, R. Agerri, I. Aldabe, A. Cybulska, M. van Erp, A. Fokkens, E. Lapa...
Resulting ECKGs
Resulting ECKGs
P. Vossen, R. Agerri, I. Aldabe, A. Cybulska, M. van Erp, A. Fokkens, E. Laparra, A. Minard, A. P. Aprosio...
ECKG Quality Evaluation
• Random sample of 100 Events from Wikinews
• 4 groups of 25 events (~260 triples), each evaluated...
Applications: Query for Occurrences per Year
Applications: Network
Applications: Interactive Visualisation
Conclusions
• ECKGs capture “who”, “what”, “where”, “when”
• Deep NLP techniques can be used to extract events from
large ...
Current & Future Work
• Adapting the tools to the humanities domain for
• Working on capturing perspectives (QuPiD2) and
s...
Play with the Wikinews ECKG at:
https://knowledgestore.fbk.eu/
Code and documentation at:
http://www.newsreader-project.eu...
NewsReader was funded by the European Union’s 7th Framework
Programme (ICT-316404)
This work is continued with support fro...
Nächste SlideShare
Wird geladen in …5
×

Building Event-Centric Knowledge Graphs from News

1.060 Aufrufe

Veröffentlicht am

Presentation given at ISWC 2016 on "Building Event-Centric Knowledge Graphs from News". The paper was published in Journal of Web Semantics: http://www.sciencedirect.com/science/article/pii/S1570826815001456

Veröffentlicht in: Technologie
  • FREE TRAINING: "How to Earn a 6-Figure Side-Income Online" ... ◆◆◆ http://ishbv.com/j1r2c/pdf
       Antworten 
    Sind Sie sicher, dass Sie …  Ja  Nein
    Ihre Nachricht erscheint hier
  • Tired of being scammed? Take advantage of a program that, actually makes you money! ■■■ https://tinyurl.com/y4urott2
       Antworten 
    Sind Sie sicher, dass Sie …  Ja  Nein
    Ihre Nachricht erscheint hier
  • Want to earn $4000/m? Of course you do. Learn how when you join today!  http://ishbv.com/ezpayjobs/pdf
       Antworten 
    Sind Sie sicher, dass Sie …  Ja  Nein
    Ihre Nachricht erscheint hier
  • Gehören Sie zu den Ersten, denen das gefällt!

Building Event-Centric Knowledge Graphs from News

  1. 1. Building Event-Centric Knowledge Graphs from News Marco Rospocher, Marieke van Erp, Piek Vossen, Antske Fokkens, Itziar Aldabe, German Rigau, Aitor Soroa, Thomas Ploeger, Tessel Bogaard http://www.newsreader-project.eu
  2. 2. Event-Centric Knowledge Graphs (ECKGs) • An ECKG represents changes in the world • Can capture long-term developments and histories of entities • Complementary to (static) encyclopaedic information found in (entity-centric) knowledge-graphs
  3. 3. Why ECKGs? • Policy makers and information specialists need to know what’s happening/what’s happened to an entity • Current de facto standard is Google/Information broker document-based search • A ‘database’ of events would make their work easier • Carried out as part of the NewsReader project Jan 2013 - Dec 2015 (EU FP7 project) • http://www.newsreader-project.eu
  4. 4. Processing Pipeline Step 1: Information Extraction at the Document Level Step 2: Cross- document event coreference opinion miner word sense disambiguation multiwords tagger syntactic parser tokenizer part-of-speech tagger named entity recognizer named entity disambiguation nominal coreference resolution semantic role labeler event coreference resolution time and date recognition Porsche family buys back 10pc stake from Qatar Descendants of the German car pioneer Ferdinand Porsche have bought back a 10pc stake in the company that bears the family name from Qatar Holding, the investment arm of the Gulf State’s sovereign wealth fund. All of the common shares in Porsche Automobil Holding SE are now held by the Porsche-Piech family, descendants of the eng- 17/06/2013 Qatar Holding sells 10% stake in Porsche to founding families Qatar Holding, the investment arm of the Gulf state’s sovereign wealth fund, has sold its 10 percent stake in Porsche SE to the luxury carmaker’s family shareholders, four years after it first invested in the firm. Qatar Holding, which owns stakes in some of the world’s largest companies, said it sold the common shares in the automaker to the Porsche and Piech families. It did not disclose the value of the transaction. @prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> . @prefix time: <http://www.w3.org/TR/owl-time#> . @prefix eso: <http://www.newsreader-project.eu/domain-ontology#> . @prefix gaf: <http://groundedannotationframework.org/gaf#> . @prefix nwrontology: <http://www.newsreader-project.eu/ontologies/> . @prefix sem: <http://semanticweb.cs.vu.nl/2009/11/sem/> . @prefix fn: <http://www.newsreader-project.eu/ontologies/framenet/> . <http://www.newsreader-project.eu/instances> { <http://www.telegraph.co.uk#ev2> a sem:Event , fn:Commerce_buy , eso:Buying ; rdfs:label "buy" , "sell"; gaf:denotedBy <http://www.telegraph.co.uk#char=15,19> , <http://english.alarabiya.net#char=14,19> . <http://dbpedia.org/resource/Porsche> rdfs:label "Porsche" , "founding family" ; gaf:denotedBy <http://www.telegraph.co.uk#char=0,7> , <http://english.alarabiya.net#char=33,40> , <http://english.alarabiya.net#char=44,61> . <http://www.newsreader-project.eu/data/cars/non-entities/10pc+stake> rdfs:label "10pc stake", "10 % stake in Porsche" ; gaf:denotedBy <http://www.telegraph.co.uk#char=25,35>, <http://english.alarabiya.net#char=20,40> . <http://dbpedia.org/resource/Qatar> <?xml version="1.0" encoding="UTF-8" standalone="yes"?> <NAF version="v3" xml:lang="en"> <nafHeader> <fileDesc creationtime="20130617"/> <public uri="5BC0-9GD1- F0JP-W2H2.xml"/> <linguisticProcessors layer="srl"> <lp name="ixa-pipe-srl-en" timestamp="2014-02-58T19:28: 32+0100" version="1.0"/> temporal relation extraction causal relation extraction factuality detection event coreference resolution <?xml version="1.0" encoding="UTF-8" standalone="yes"?> <NAF version="v3" xml:lang="en"> <nafHeader> <fileDesc creationtime="20130617"/> <public uri="5BC0-9GD1- F0JP-W2H2.xml"/> <linguisticProcessors layer="srl"> <lp name="ixa-pipe-srl-en" timestamp="2014-02-58T19:28: 32+0100" version="1.0"/>
  5. 5. NLP Pipeline • State-of-the-art pipeline • English, Spanish, Italian & Dutch • ‘Black box’ setup and individual modules available from: http://www.newsreader-project.eu/results/software/ • All modules take NAF as input and output NAF
  6. 6. NLP Pipeline output
  7. 7. Cross-Lingual Event Extraction
  8. 8. Cross-Lingual Event Extraction
  9. 9. Text2RDF
  10. 10. Event and Situation Ontology (ESO)
  11. 11. Event and Situation Ontology (ESO) See also: P. Vossen, R. Agerri, I. Aldabe, A. Cybulska, M. van Erp, A. Fokkens, E. Laparra, A. Minard, A. P. Aprosio, G. Rigau, M. Rospocher, and R. Segers, “NewsReader: How Semantic Web Helps Natural Language Processing Helps Semantic Web,” Special issue Knowledge-Based Systems, Elsevier, 2016.
  12. 12. Resulting ECKGs
  13. 13. Resulting ECKGs P. Vossen, R. Agerri, I. Aldabe, A. Cybulska, M. van Erp, A. Fokkens, E. Laparra, A. Minard, A. P. Aprosio, G. Rigau, M. Rospocher, and R. Segers, “NewsReader: How Semantic Web Helps Natural Language Processing Helps Semantic Web,” Special issue Knowledge-Based Systems, Elsevier, 2016.
  14. 14. ECKG Quality Evaluation • Random sample of 100 Events from Wikinews • 4 groups of 25 events (~260 triples), each evaluated by 2 human raters • Strict co-reference evaluation; if an event had 4 mentions of which 3 were correct, all coreference triples were considered incorrect
  15. 15. Applications: Query for Occurrences per Year
  16. 16. Applications: Network
  17. 17. Applications: Interactive Visualisation
  18. 18. Conclusions • ECKGs capture “who”, “what”, “where”, “when” • Deep NLP techniques can be used to extract events from large amounts of text • Distinguishing between mentions and instances allows for cross-lingual event extraction through semantic layer (no MT needed) • ECKGs were evaluated and tested in hackathons
  19. 19. Current & Future Work • Adapting the tools to the humanities domain for • Working on capturing perspectives (QuPiD2) and storylines (ULM-3) • Your domain/project? Get in touch: marieke.van.erp@vu.nl
  20. 20. Play with the Wikinews ECKG at: https://knowledgestore.fbk.eu/ Code and documentation at: http://www.newsreader-project.eu/results/software/
  21. 21. NewsReader was funded by the European Union’s 7th Framework Programme (ICT-316404) This work is continued with support from the CLARIAH-CORE project financed by NWO

×