3 ms·
As someone who specialized in this area for years in social media specifically, I have never seen evidence you can get past 80% accuracy with any known sentimen
by lrvick 4y ago
As someone who specialized in this area for years in social media specifically, I have never seen evidence you can get past 80% accuracy with any known sentiment analysis method.
Brute force stats like Naive Bayes is still as good as it gets on platforms like Twitter where speech patterns are too short and irregular for traditional NLP to be very useful.
- elgenioso 4y agoEven non-traditional NLP won't work. There is no way to algorithmically enforce "good" speech. I agree that the character limitations don't help and the 280 character limitation is the main reason Twitter is such a mess. There are no coherent long form texts generated by AI models but 280 characters is very easy because people will give the author the benefit of the doubt and interpolate the missing context.
- MichaelCollins 4y ago> 80% accuracy Sentiment analysis is way worse than that in adversarial scenarios, where a human is trying to outwit the machine. My prediction for this whole thing: Elon Musk [again] gets swept up by AI hype, tries to replace humans with software, and it doesn't work. Like at Tesla, so at Twitter.