[Question] Wikipedia pre-training dataset about ernie HOT 5 CLOSED

As the Wikipedia pretraining data is quite large, we will not release the full data. However, we will give some instructions about how to construct the training data recently. And, we have released the Wikidata embedding here.

from ernie.

@zzy14 Thanks for releasing your model and code! I have a follow up question regarding the wikipedia pretraining data: what is the entity linking tool you used to obtain aligned data? From the example code, it seems that the TAGME API is slow. Thanks!

from ernie.

@ledw In Wikipedia corpus, there are human-annotated alignments from the anchors.

from ernie.

@zzy14 Thanks for your reply. That makes a lot of sense. Any chance that you could also release the code for preprocessing the wikipedia corpus?

from ernie.

I am wondering when we can get instructions about how to construct the training data using Wikipedia?

from ernie.