Tom, thanks for the link. I've been to the site in the past. I just revisited, and entered "Sarah Palin" in the search box. It returned 50 tweets, labeled 39 as neutral, 5 positive and 6 negative. I checked the actual tweets, and here's what I found:
- Neutral: I only found one (maybe). The other 38 were actually pretty negative. - Positive: only one was positive, the other four were sarcasm. - Negative: you did well here, 5 were negative, and one was actually positive. So, the accuracy level on this test is 6 or 7 out of 50, or anywhere from 12% to 14%. Given the state of the art in computing power, I think we're still years away from NLP and Machine Learning, being able to properly process sarcasm, double entendre, backhanded compliments, turns of word, etc. So that may work in a limited fashion, and for certain topics, where the format and style are controlled, but not when it's free for all. On Jul 23, 1:28 pm, tomz <tom.z.z...@gmail.com> wrote: > http://tweetsentiments.comuses Machine Learning and NLP for sentiment > analysis. > No published API access yet(some already available), but it's on the > road map. > Currently sacrificing some precision for speed, but we will focus on > improving precision in the near future.
