How to build the feature vector from sentence for intent classification in NLU?

Question

I am trying to develop a NLU (natural language understanding) engine which interprets human language utterance to intent and slots. After some searching, I found this very useful question for NLU novice like me: How to proceed with NLP task for recognizing intent and slots

In the answer, @darshan says:

...use a classifier to do the same using feature vector formulated from the input sentence.

I know that after transforming the sentence to a feature vector will be very useful for classification task. But how should we get such a feature vector?

For instance, the sentences: What's the weather today? What's it like out? How's the weather? all should be classified to a weather_query intent. But how to make these sentences into numerical feature vectors?

Answer 1

I suggest you build a classifier for each intent and then use that to find any given intent in a sentence. For starters, work using a rule based system such as using a keywords or regex. Then once you've enough sentences go forward with the machine learning classifier and keep improving it.