I would suspect that instead of word frequency, they likely use TF-IDF (or even better, BM25[1]) or LDA [2] to identify relevant keywords to highlight.
As an aside - have you read about the supposed NLP used by Obama in his speech cadence and delivery? I think its quite fascinating... just google Obama NLP
> Overall there’s too much information that can’t really get stuffed into 140 characters which is quite a sham
You could have a "verbose" mode (e.g. put -v at the end of the tweet) which abbreviates everything to one or two characters and omits the best review phrase.
10 comments
[ 4.0 ms ] story [ 27.6 ms ] threadYelp has also started doing something along these lines selecting quotes from reviews in which a word that occurs frequently has been found.
1. https://en.wikipedia.org/wiki/Okapi_BM25
2. https://en.wikipedia.org/wiki/Latent_Dirichlet_allocation
As an aside - have you read about the supposed NLP used by Obama in his speech cadence and delivery? I think its quite fascinating... just google Obama NLP
You could have a "verbose" mode (e.g. put -v at the end of the tweet) which abbreviates everything to one or two characters and omits the best review phrase.