We're going to to perform some tasks concerning sentiment classification of tweets?
Which features do you think should be taken into account?
For sentiment classification of tweets, I have found that unigram, bigram features selected through information gain measure are effective. You may also include Hash tag information as a feature also.
Dear Yaakov,
an article that could interest you?
http://www.scirp.org/journal/PaperInformation.aspx?PaperID=61354&utm_campaign=Paper&utm_source=e_nl&utm_medium=jdaip_20151130_wangjiasi_58145
Best regards,
Ludovic
Hey! You should have a look at the proceedings of SemEval (http://alt.qcri.org/semeval2015/cdrom/), maybe also consider 2014 and 2013. The participants investigated a lot of features for Twitter Sentiment Classification there.
In case of tweets that are generally grammatically incorrect sentences you have to consider the lexical specifics of tweets -- they can contain repetitions, inexact wordings, corrections etc. For classification I would consider stylometric features and probabilistic classifiers such as NB (see below).
Consider social aspects of the language of tweets' authors that can convey their sentiment in very different manner. For example, particular authors can mean a tweet as a positive joke and others can understand it as nasty rudeness.
Thesis Opinion Mining and Lexical Affect Sensing
- Punctuation (? ! ..)
- Smileys
- Tweet length
- Onomatopoeia (hhhhh, hahahaha, hehehe, jjjjjj, boom, pffff, argh,...)
There seems to be a related study.
No better link available.
https://www.newscientist.com/article/dn25623-tweets-map-the-worlds-emotional-response-in-real-time/
Maybe relevant as well:
https://medium.com/the-physics-arxiv-blog/text-analyser-reveals-emotional-temperature-of-novels-and-fairy-tales-92f0800cb724
Regards,
Joachim
I think whatever you do, it should include semantic analysis. Maybe you can use a conceptual ontology for languaje processing and also sentimental status recognition. Conceptual analysis can help you in understanding human behavior. Good luck!!
You can use
lexical feature as unigram, bigram, trigram
pragmatic feature as smilies, emoticons, replies
hyperbole feature as intensifier - adjective and adverb
punctuation- !, ??
quotes- " ", ' '
Using these feature set, easily can find sentiment of a tweet.
We done some seven variations test from 500 human raters on seven variations. This is the first and second paper about it, I am producing another journal writing and soon to be published, will update you if you interested. how about a cross country join research?
https://www.researchgate.net/publication/282239839_Exploring_fine-grained_sentiment_values_in_online_product_reviews
https://www.researchgate.net/publication/284390575_Sentiment_Analysis_Tools_Should_Take_Account_of_the_Number_of_Exclamation_Marks%21%21%21
Conference Paper Exploring fine-grained sentiment values in online product reviews
Conference Paper Sentiment Analysis Tools Should Take Account of the Number o...
according to me in sentiment analysis words play important role,so it would be good to know the categorie of word
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science
- Materials
- Inorganic Materials
- Metals
- Alloys
- Steel