I would like to obtain word embeddings (vector representation of words).

Do you think it is better to try to directly obtain a word embedding using some neural network approach or is it better to learn character embeddings then, use them to build word embeddings?

  • 1
    Use the words directly, not the characters. Yann LeCun did a text prediction study years ago where he found much higher accuracy using characters rather than words. That was the only evidence; other studies showed words were better.
    – Mitch
    May 26 '17 at 13:05

I recommend that you read this article by Ling et al:

Author = {Wang Ling and Chris Dyer and Alan Black and Isabel Trancoso},
Title = {Two/too simple adaptations of word2vec for syntax problems},
Booktitle = {Proceedings of NAACL},
Year = {2015},

It contains a guidance on doing word embeddings properly.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.