Is there a way to measure the performance of a Word2Vec model?

Christos_Karapapas

I am using the word2vec extension to train a model for polynomial text classification.
None of the standard Performance operators seem to "stack" with the word2vec model (RMWord2VecModel).

Is there any way that I could measure the performance of the model on the training dataset?

MartinLiebig

Hi @Christos_Karapapas ,

good question! Usually Word2Vec is either evaluated empircally (close words should be "synonyms", close in meaning) or by using down stream tasks like Entity Recognition.

Cheers,

Martin

Christos_Karapapas

So, if I understand it correctly it's just another way of finding synonyms just like when grouping words by their root (lemmas)?

And if so why it exports a model? To be used in later processes with different datasets and still be able to find the lemma of a word?

Telcontar120

Yes, it is more of a processing model than a predictive model.  Most unsupervised ML approaches don't have th same kind of performance measurement as you are used to from standard predictive models.