On Chatbot

People think my chatbot is limited to dependency parsers. Graphs allow so much more than that. Combined with other techniques, they will beat a lot of supervised systems at a fraction of time and cost. I can also easily incorporate semantic parsers into the graph solution.

If you think about it, supervised learning doesnt make sense for NLP. Once you have taught the system how to process sentences using parsers, you shouldn't need to train it again and again. The closest thing is transfer learning. My  system mimics the human language processing system in a close way.

The main components of chatbot are:
1. Sequence modelling
2. Classification

Parsers and graphs do sequence modelling. Linguistics can also be used for sequences. Classification also handled using joins. Rest one can do tfidf or multiclass multilabel classification using ngrams. Add LSA for handling variety.

I am the king.