chatterbot and machine learning

[brightening-eyes]

hi,
In this issue, I would like to share my concerns about the use of chatterbot with deep learning stuff.
you can comment as you wish, and when I have the time, I will do my best to implement and/or provide a way for others to implement them

• word embedding and sentence similarity
  instead of spacy, in the comparisons, we can use a deep learning model (provided by user) to perform text similarity (by using pre-trained word vectors). then no need for spacy, but instead, a module for deep learning like tensorflow, pytorch, mxnet etc is required.
  the downside however, is that the user should bring their pre-trained model here, which is reasonable because many pretrained models are used everyday.
  additionally, a way to preprocess the sentence for model should be available here, as each model is different.
  with this addition, best_match adapter automatically works with machine/deep learning models.
• instead of nltk in time logic adapter, we can use scikit-learn for classification
  this removes nltk, but adds scikit-learn, which seems more reasonable.
  lower in size, no need to download nltk's dataset, etc.
  also we are using spacy instead of nltk. aren't we?