Emotion analysis on Twitter for past 5 years
Emotion analysis is a new research direction in computer science in these years, what it does is expressing emotions and feelings of people with data. The means to do emotion analysis include natural language processing, statistical and signal processing. The platforms which are most suitable for emotion analysis are also microblogging like social networking platform, Twitter is one good example.
Computing Laboratory of the University of Vermont project Hedonometer conducted emotion analysis for the tens of millions tweets every day in the past five years and recorded the results. Hedonometer analyze emotions of people mainly through natural language processing, for example, it will find some keywords reflecting positive emotions or negative emotions in a tweet.
We can see from the figure above that the happiness is the highest is on 25 December every year. Other happy days including New Year's Day, Thanksgiving, Valentine's Day. For each week, Saturday is the most happy day and Tuesday is the least happy day.
But since 2008, the overall degree of happiness on Twitter has been declining. This may be because the U.S. financial crisis in 2008, and after that the continuing economic downturn. We can see the least happy day of April 2013 is on the 15th of the month which was the the Boston Marathon explosion occurring day.
But the algorithm used by Hedonometer is obviously not perfect, On May 2, 2011 the United States announced that Osama bin Laden was killed that day, but Hedonometer mistakenly believed that this day of happiness was very low because of some negative words such as the "death", "kill" appeared frequently on Twitter. It is one of AI Complete algorithm problems to fully understand the natural language of human beings.
If you are interested in the detailed algorithm used by Hedonometer.You can refer this article.
Source : http://www.36kr.com/p/203074.html