Using node.js and natural language processing to handle multiple word phrases

Question

I'm using the very cool natural library for node.js.

I'm trying to train my classifier to match the phrase user experience. My issue is if I do something like this:

classifier.addDocument(['user experience'], 'ux');

It doesn't match 2 word phrases, I believe because it tokenizes the words. If I do something like this:

classifier.addDocument(['user', 'experience'], 'ux');

It works like I want it to, but my issue is, I don't want to just match on the word user because an article could mention include the word user multiple times and it would potentially have nothing to do with user experience, which would lead to inaccurate classifications. So, my question is how does one match 2 or more word phrases using NLP?

Thanks for you help in advance.

Answer 1

You should take a look at n-grams, specifically in this case it's called a bigram, a sequence of two tokens. https://github.com/NaturalNode/natural#bigrams

I haven't used that particular library (don't think nodejs is the best language for NLP, it's still in its early stage and I'd suggest you use a more mature library(NLTK)/language(python) for NLP. Though I guess it's fine just for testing or some small project).

Anyway, judging from the manual, you could maybe do something like

classifier.addDocument([['user', 'experience']], 'ux');

Add brackets for each sequence you wish to add together.