Pretrained word embedding

Question

Pretrained word embedding

HAWLYQ opened this issue 6 years ago · comments

@glample Hi, in your paper, it was mentioned that you observed significant improvements using pretrained word embeddings. Could you share your pretrained English word embedding with me?

Guillaume Lample · Answer 1 · Thu Oct 11 2018 17:05:04 GMT+0800 (China Standard Time)

Hi,

Yes, you can find them here: #44

Enrico Versino · Answer 2 · Fri May 24 2019 05:05:37 GMT+0800 (China Standard Time)

Hi.
I've tested your embeddings and were great. Could you share pretarained italian and french word embeddings?

Guillaume Lample · Answer 3 · Fri May 24 2019 09:11:10 GMT+0800 (China Standard Time)

Hi,

I never trained embeddings on these languages, but you can have a look at the fastText embeddings, they should work very well for NER: https://fasttext.cc/docs/en/crawl-vectors.html

M Saiful Bari · Answer 4 · Fri May 24 2019 09:15:52 GMT+0800 (China Standard Time)

@glample
I recently used your embeddings.
your skip-n-gram embeddings given here are better than fasttext common-crawl vectors except the vectors for dutch.
in case of fasttext, dutch pretrained embeddings gives around 85.75f1 score where skip-n-gram gives around 81.**

Enrico Versino · Answer 5 · Fri May 24 2019 15:57:29 GMT+0800 (China Standard Time)

Thanks From: Guillaume Lample Sent: Friday, May 24, 2019 3:11 AM To: glample/tagger Cc: Enrico Versino ; Manual Subject: Re: [glample/tagger] Pretrained word embedding (#83) Hi, I never trained embeddings on these languages, but you can have a look at the fastText embeddings, they should work very well for NER: https://fasttext.cc/docs/en/crawl-vectors.html — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.