3. State of the art
Zemanta and Alchemy API Sil et al.
DeMartini et al.
Boyce-Codd
Normal Form
4. Preliminary Results
• Can we apply state-of-the-art algorithms for Entity Linking with other
Knowledge Bases?
Knowledge Base F-Score
0.87
0.54
me 32
(Andrea Moro, Alessandro Raganato, Roberto Navigli. Entity Linking meets Word
Sense Disambiguation: a Unified Approach. Transactions of the Association for
Computational Linguistics, v. 2, 2014.)
5. Research Questions
• What are the textual and Knowledge Base features most relevant for Entity
Linking?
mentio
n
mentio verb ?
n
Marco Nanini performed very well as Alberto in Copacabana movie.
• Is it feasible to achieve comparable performance to Knowledge Base specific
approaches?
Context specific modules
Marco Nanini
Copacabana (2001) Copacabana
Alberto
There are different Copacabanas.
Years of good results (NER, EL with Wikipedia and Wikipedia-based KBs in Named Entities)
There is no single KB that contains the whole knowledge in the world. The use of multiple ones is desired.
This is not new!
But the state-of-art in Entity Linking with a single KB has not been used..
Do it myself.
Nice results. Everybody likes nice results.
Bad results exist. What to do with them? Why is the system bad? (The results are just numbers)
Textual (local/global) and KB features (entities and relations).
Can I have nice results? I also want them. Maybe using context specific modules.
Dictionary of Names for Mention Recognition
Ontology Modularization for Candidate Selection
Collective Inference for Disambiguation
Annotated Corpora / Coverage for the Mention Recognition > Presence of all candidates in the same module > Evaluation of textual and KB features as well as Collective Inference Method
- / Coverage / Coverage / Precision and Recall
Do not follow that path! I tell you why!
Encourage the community to work KBs other than cross-domain ones. (Reputation Management, Product/Brand detection, and the WEB need this!)