4 ms·
nice work ! btw, personally I am not feeling that bad about some negative news. we need to know the whole picture, right? haha, my 2c
by nittanymount 3y ago
nice work !
btw, personally I am not feeling that bad about some negative news. we need to know the whole picture, right? haha, my 2c
- PaulHoule 3y agoOverall I don’t think HN has a lot of negative posts but it does have negative comments. My RSS reader YOShInOn has a model that predicts if a headline will get a lot of votes, things that score very high on that one tend to be “clickbait” articles that create “knock down drag out” discussions. (e.g. anything about low birth rates in Japan) I think he is using the Vader model from NLTK which is claimed to be pretty good although I am always singing “NLTK is a joke” because most of the models in it are in the “nice try but no cigar” category. Sentiment models do need to be tuned to the domain. It is easy to get data sets for product reviews for instance, but there is a very definite language people use to talk about the stock market or that use while trolling in social media. I want a sentiment model trained on “angry toots” from Mastodon which would probably be about an FTE-week worth of work to build a training set for (e.g. the time it takes to play a moderate sized video game) but looking at 5000 angry toots might just kill me so I haven’t done it.