TURL embeddings

[albertus-andito]

Hello again,

I see that you're including the generated TURL embeddings along the datasets.

I'm just wondering if you could also share/explain the code that generates those embeddings?
In particular, because TURL is very entity-heavy and the ALITE datasets are not, how did you decide which columns on the ALITE tables should be the entities and did you just assign randomly unique entity ID to each value in the column?

Thanks

[aamodkh]

Hi Albertus,

Please find the code to generate different embeddings for the new dataset here: https://github.com/northeastern-datalab/alite/tree/main/codes/align_utilities.
Other embeddings such as Sentence BERT and RoBERTa are also available.
You can find the config details on the TURL GitHub repository.
Also, TURL entity_vocab.txt file is provided by the authors here (inside data folder).