Add another NLP model

Quotes are proposed dependent on the sentiment scores of the `nltk` library.

It seems like it needs some improvement to return quotes which are more similar to the sentence the user typed in.

Start by tracing how the sentence entered by the user is scored with the `nltk` library and how that score selects quotes. Review the current sentiment-based approach before deciding how another NLP model could improve similarity. Done means the selected quotes better match the user's sentence, with the comparison criteria made explicit.

Written by the indexing model from the issue text.