Risto Hinno
Simple app made as a part of Data Science Capstone.
Idea. Use existing data (texts) to build a model that predicts what is next word in the text (more information).
Model uses:
3 482 415 three-grams (example: “my name is”)
1 492 091 two-grams (example “I am”)
3 one-grams
Simple backoff model using n-grams:
Use plot to monitor overall and each guess accuracy:
This functionality could be turned off from the input panel
Expected overall accuracy is 25%:
1st guess 15%
2nd guess 5%
3rd guess 5%
In reality it might be lower if data is different from data used to build the model.