问题
I'm trying to implement Sentiments Analysis functionality and looking for useful features which can be extracted from tweet messages.The features which I have in my mind for now are:
- Sentiment words
- Emotion icons
- Exclamation marks
- Negation words
- Intensity words(very,really etc)
Is there any other useful features for this task? My goal is not only detect that tweet is positive or negative but also I need to detect level of positivity or negativity(let say in a scale from 0 to 100). Any inputs or references to printed papers are very welcome.
Thanks.
回答1:
Others that may be useful are:
- elongated words (eg. goooood)
- unigrams and bigrams of every word (particularly if you have a large corpus)
Regarding references: This tutorial by Christopher Potts is very good and to the point: http://sentiment.christopherpotts.net/
Other papers:
- Twitter as a Corpus for Sentiment Analysis and Opinion Mining. Alexander Pak, Patrick Paroubek
- Twitter Sentiment Classification using Distant Supervision. Go et al. 2009.
- Robust Sentiment Detection on Twitter from Biased and Noisy Data. Barbosa and Feng. 2010.
- Sentiment strength detection in short informal text. Thelwall et al. (2010). JAIST
回答2:
If I post really good news on twitter, a lot of people might start publicly congratulating me.
So If I post X, and then get a lot of 'Congrats' tweets from other people, then X is probably positive.
In general, the type and frequency of people who retweet my tweet might have something to do with its inherent sentiment.
回答3:
I would suggest the following articles:
- Belgian elections, June 13, 2010 - Twitter opinion mining, http://www.clips.ua.ac.be/pages/pattern-examples-elections
- 100 days of web mining, http://www.clips.ua.ac.be/pages/pattern-examples-100days
来源:https://stackoverflow.com/questions/8322609/twitter-sentiments-analysis-useful-features