TREC-50 SOTA

[hossein-amirkhani]

The TREC-50 SOTA is changed by the "Universal Sentence Encoder" paper.

[sebastianruder]

I think the results from USE are for TREC-6 instead of TREC-50 as they use the dataset prepared by Conneau et al. (2017) who use TREC-6.

[hossein-amirkhani]

But they mention "fine grained question classification" to introduce the TREC task. By "Fine-grained" I infer TREC-50.

[sebastianruder]

As I said, they use the data from Conneau et al. (2017) who do TREC-6 with 6 classes. This is already more "fine-grained" than binary classification. If they do TREC-50, then there's a mistake in the paper.

[hossein-amirkhani]

So, if you are sure, I close this PR and creatre another PR for TREC-6.