I have a decent understanding of word embeddings (at its core, one can think of a word being converted into a vector of, say, 100 dimensions, and each dimension given a particular value... this allows to do math with the words, also it makes the training sets to be non-sparse...)

But today something came to my mind, what about punctuation symbols such as , . () ? ! ... ?

They do have a huge impact on the meaning of sentences and, like words, the position and context in which they are used is relevant.

So the question is, how should this be modeled? are pretrained sets like GloVe including punctuation symbols? should I simply remove punctuation symbols from text?


Punctuation such as , . () ? ! ... ? are all included in pretrained word vectors such as Glove. However, it is common to find phrases, abbreviations, and misspellings that are not contained within pretrained vectors. What to do with those phrases is highly dependent on the goals of the analysis and context of the text.

| improve this answer | |
  • $\begingroup$ agree with @J_Heads $\endgroup$ – Sunil Jan 23 '19 at 3:45
  • $\begingroup$ I would also add that it depends on the pretrained model you use, for example in Google's pretrained Word2Vec they have all the variations of the words including the misspelled ones and even non words tokens. In general it depends on the decision made by the one that created and trained the model, whether he did some aggressive pre processing or not for example... $\endgroup$ – Oleg Feb 3 '19 at 23:59

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.