NxT_presentation

• It's a predictive words app using a NLP model.
• The model used is very simple, it's just a stupid backoff model training with a lot of part of texts provided.
• I selected via a binomial way, 10% of all of texts merged, and I created ngrams 1, 2, 3 and 4.
• I repeated the same process again and again in changing the seed. Finally I merged every ngram with the same size.

The model is basic, it takes the last 3 words and goes to check if there is something in the ngram. If not, it checks the ngram smaller etc. If nothing is found, it give the 5 words most use. The performance accuracy was of 72% on the test texts used. It surely possible to improve it, using a better method, but also heavier in computing. I made the choice for the rapidy of computing, to make it easier to use it. Furthermore, the model is training with the user, and so improve itself.

You just have to launch the app, and enter your word(s) in the input zone. Be careful, if you tape a space in finishing, the app will wait a new word and so will search nothing. You have to finish on by a word.

I hope you will have as much fun using it as I had to build it ! Thank you for reading me to the end.