Ad verba per numeros

Wading through mood
HOT!, Artículos, Doctorado y máster
Tuesday, June 16, 2009, 03:10 PM
Lately I'm pretty interested in sentiment analysis a.k.a. opinion mining. Obviously, I'm not the only one and there exist several applications trying to do some of that stuff over user generated content (e.g. twitrratr). Needless to say, performance is quite limited and there is still much room for improvement (which, from my point of view, are good news).

A couple of weeks ago I suggested to a group of students to develop a naïve subjectivity analyzer based on the Subjectivity Lexicon compiled by Wilson, Wiebe and Hoffmann (2005).

Some of them have developed pretty amazing apps including a couple of Twitrratr clones, another one to give scores to movies based on their reviews and a really inspiring tool by Diego Guerra to "detect" the users mood from the songs they listen to at last.fm.

To do that he employs WordNet Affect, to be precise the subset provided for SemEval-2007, and, of course, a pinch of "magic" :)

Here you have a couple of pie charts showing the different moods from three different last.fm users:




Have I aroused your curiosity? Great! Here you have three papers you can find of interest:And just one last little thing: if you are a Computer Science student at the University of Oviedo, you are looking for an interesting final year project, and have found this intriguing, just give me a buzz.

P.S. Another useful/interesting/related resource: We feel fine (home page, API, data).


Next