Review of baselines and state of the art comparing methods

[fedenanni]

We need two state of the art comparisons:

End-to-end pipeline (so a comparison with a tool that given a string predicts the entities):

• easy-to-use baseline system: https://github.com/marcocor/tagme-python
• state-of-the-art: https://github.com/informagi/REL selected from NLP Progress, it seems super easy to use out of the box.

Only disambiguation approach (so given a string and a set of candidates):

• maybe this one, if it's easy to use: https://github.com/openai/deeptype

For both we could use them directly out of the box or the same pipeline with our own language models if possible.

[mcollardanuy]

From #9:

We should read this survey, which covers most typical EL baselines and see which ones we want to implement: http://www.projectsgoal.com/download_projects/cloud-computing/cloud-computing-projects-GCC00111.pdf

[mcollardanuy]

From #19:

• Entity linking:
  • https://github.com/nicola-decao/efficient-autoregressive-EL, using candidates from https://github.com/masha-p/PPRforNED
  • https://aclanthology.org/2021.emnlp-main.709.pdf
  • https://aclanthology.org/2021.emnlp-main.820.pdf
  • https://arxiv.org/pdf/1909.00426.pdf
• Toponym resolution:
  • ...
• Graph-enhanced LM: https://github.com/snap-stanford/GreaseLM

Multilingual entity linking:

https://link.springer.com/content/pdf/10.1007/s00799-021-00319-6.pdf

GeoBert:

https://aclanthology.org/2021.vardial-1.16.pdf