Hacker News new | ask | show | jobs
by adhsu01 3898 days ago
Nice! Congratulations on putting this out there.

How does this work? What did you use as training data? Do you parse the sentence in any way?

For example, "Hey! It was so so great to see you." gets a 46/100, and 'so' is listed as an impolite word, which is understandable in some contexts but not this one.

1 comments

Thanks. Yes, we built our own training set.

Catching all the different senses of the word is a hard problem. There's always something to improve.