I am using two different models, Latent Dirichlet Allocation (LDA) and WordVec, to create feature vectors for document classification. The output of the LDA model is a probability vector, i.e. the components are all non-negative and sum to 1. The vectors derived from the WordVec algorithm are not probabilities, i.e. they are not non-negative and do not sum to 1. I am wondering how I can combine these two vectors. One thought I have is to softmax the WordVec vectors and then do the combination (which will be a concatenation). Otherwise, I'm not sure how to properly normalize the concatenated vectors.

  • $\begingroup$ It would be very helpful to understand what the ultimate goal is. Why do you need the normalization? For example, if you plan on using the concatenated vectors to further feed into some neural network, then the details are irrelevant, just input the concatenation. $\endgroup$ – Alex R. Apr 25 '17 at 18:41
  • $\begingroup$ I plan to use them as feature vectors for a logistic regression model. $\endgroup$ – thecity2 Apr 25 '17 at 18:42

Your Answer

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

Browse other questions tagged or ask your own question.