search

ccsasuke / man

Multinomial Adversarial Networks for Multi-Domain Text Classification (NAACL 2018)
https://arxiv.org/abs/1802.05694
MIT License
80 stars 28 forks source link

Source of Word Vectors #2

Closed Humanity123 closed 6 years ago

Humanity123 commented 6 years ago

Experiment 3 uses pre trained word vectors of 100 dimension. It has not been mentioned in the paper or the repository the source of the embeddings. Can you please briefly explain the method used for training the embeddings? Moreover does the choice of embeddings affect the results?

ccsasuke commented 6 years ago

Hi,

We actually (briefly) mentioned how the embeddings are obtained in the paper:

We use word2vec word embeddings (Mikolov et al., 2013) trained on a bunch of unlabeled raw Amazon reviews (Blitzer et al., 2007).

We haven't experimented with alternative embedding choices, so I don't have an answer for that question at this moment.