Home

This Wiki is mainly for making notes of how I tackle NLU through Snips NLU Engine.

How I get here

Started out going down the path as a pure NLP approach - remove punctuations, tokenization, lemmization, stemming, then extract weighted features through TF-IDF.
I then throw these featuers into Doc2Vec and thought I can get a reasonable classification.
However, it fails to work for the need of identifying "intents" behind natural languages and goes heavily toward the frequency of a word being used. This will definitely help with identifying keywords in a broader perspective, but not so much to a broader spectrum like the intention behind it.
This is the moment I dicovered Snips NLU ( and a similar package called Rasa NLU. Open sources specialized for contexual and natural language understanding!

Details on each sections

• p2.py - Outlier Seeker for NLU Problem -> Overview + Approach
• Snips NLU -> Work being done

There's an additional folder : other that is meant for an unrelated script - p1.py.

In a nutshell, it is a simple function to find the max score(number) in a list = count(element in list > this score).

• See p1.py for more details.