News for hackers. Stuff that matters.

A comment on Word2Vec Received 'Strong Reject' Four Times at ICLR2013. Parent comment

Re:Word2Vec Received 'Strong Reject' Four Times at ICLR2013 by jncfhnb on Monday December 18, 2023 @12:38PM (#38685614)
To be fair I have some memories of the papers and surrounding tech being really bad. The popular implementations didn’t actually do what the papers said and the tech wasn’t great for anything beyond word level comparisons. You got some juice doing tf idf weighting of specific words but then tf idf weighted bag of words was similarly powerful.

Cosine similarity of the sum of different word vectors sounds soooo dumb nowadays imo

[ Parent | on HN ]
Re:Word2Vec Received 'Strong Reject' Four Times at ICLR2013 by derangedHorse on Tuesday December 19, 2023 @12:07PM (#38698249)
Cosine similarity is like a very primitive version of word2vec. They both work off the idea of context being the key to the derivation of a token's semantic meaning. Cosine similarity is not very useful in itself but it's helpful for understanding the progression in the space
[ Parent | on HN ]