We want to be able to put sentences into categories by user intents. Intents are a generic mechanism that combine multiple individual examples into one semantic umbrella. For example, hi, hey, good morning, and wassup! are all examples of the _greeting_ intent.
Using greeting as an input, the backend logic can then determine how to respond to the user.
There are many ways we could combine word vectors to represent a sentence, but again we're going to do the simplest thing possible: add them up.
This is definitely a less-than-ideal solution, but works in practice because of the simple, unsupervised approach we use with this:
def sum_vecs(embed,text): tokens = text.split(' ') vec = np.zeros(embed.vector_size) for idx, ...
