Datasets ?

[Khalife]

Good afternoon,

And thanks again for making available the source code of your project.
Do you have an idea about when the corresponding datasets will be available ?

Thanks in advance,

Sammy Khalife

[octavian-ganea]

Hi. The indexes have a few tens of GB, so we can send them on demand to people that are interested. Please contact us (e-mails in the paper) to see how can we do this transfer.

[Khalife]

Hi, Thanks for your answer.
I just sent a mail about this.

I still have the following questions in mind :
-Have you annotated every dataset in your list (CoNLL-AIDA, AQUAINT, MSNCB, ACE04) ?
-Can I find some of them on the web that are already annotated?
-If I'd like to train on sufficiently annotated dataset (not every one that you use), which file source code should I modify accordingly?

[octavian-ganea]

AQUAINT, MSNBC and ACE04 datasets can be obtained from http://webdocs.cs.ualberta.ca/~denilson/data/deos14_ualberta_experiments.tgz . The rest is handled during evaluation by the method in eval/datasets/AQUAINT_MSNBC_ACE04.scala .

The AIDA datasets are not public, one needs to get the license for them. Using this license, a text file with entity annotations is generated and this can be used with PBOH as shown here : eval/datasets/AIDA.scala . Just the annotations, without the full documents, can be obtained from here : https://www.mpi-inf.mpg.de/departments/databases-and-information-systems/research/yago-naga/aida/downloads/