Source of training queries for some BEIR datasets

#2
by ftvalentini - opened

I have some questions regarding the origin of the training queries used for these datasets, which have no training queries in BEIR:

  • quora: 10k test, 5k dev queries in beir -- bge-full-data has 60202 queries.
  • scidocsrr: 1k test queries in beir -- bge-full-data has 12654 queries.
  • arguana: 1406 test queries in beir -- bge-full-data has 3101 queries.

Where do these train queries come from?

Also for nli dataset: what is the source dataset?

Thank you so much for making such a valuable dataset available!

Sign up or log in to comment