3. Related work
1: Event registry: learning about world events from news, Leban, Gregor and Fortuna, Blaz and Brank, Janez and Grobelnik, Marko, 2014
2: The Concentric Nature of News Semantic Snapshots: Knowledge Extraction for Semantic Annotation of News Items, García, José Luis Redondo, Giuseppe Rizzo, and
Raphaël Troncy, 2015
● Event Registry1
● The NSS model2
○ Core
○ Crust
4. Methodology
1: The Concentric Nature of News Semantic Snapshots: Knowledge Extraction for Semantic Annotation of News Items, García, José Luis Redondo, Giuseppe Rizzo, and
Raphaël Troncy, 2015
5. Data
● ECB+ dataset2
annotated by experts
2 : Guidelines for ECB+ annotation of events and their coreference, Cybulska, Agata and Vossen, Piek, 2014
Annotation example
.. another person in 1999 while she visited him in prison .
Relations
Causal relation : pulled on → fleeing
Temporal relation : Monday → pulls on
7. Main Experiment
● Approach to solve problems
○ Lemmatization/stemming
○ Similarity comparison
○ TF-IDF
○ Considering temporal and causal relations between events
○ Core and Crust scores
■ Core: Relations with all entities in Core
■ Crust: Relation with an entity in Core
10. Discussion & Conclusion
● Corereference
● Limited scores for actors, locations and time periods
○ Better NSS models
○ Change of scores for events
Research Question:
Can we determine a relevance score of events and its participating entities in news articles?
Answer:
Yes, but…
11. Thank you for listening!
Quinten van Langen
q.i.van.langen@student.vu.nl
Oana Inel