Sentiment Analysis

Twitter has played a prominent role in the political and social changes taking place in the Middle East. We are exploring how to automatically classifying the political sentiment of tweets, so that we can better quantify political feelings of people around the world and track their changes over time.

Questions that we are examining include how do Iranians using Twitter feel about the UN sanctions? Their country's nuclear weapons program? Their own government? The United States?

Sentiment analysis with Twitter is harder due to the short text and use of jargon or abbreviations. Major work has been done with movie or automobile reviews, but in these cases the sentiment is usually easier to detect. When considering the Middle East, opinions are usually expressed in more subtle or indirect ways. In addition, many of the tools used for analyzing English, such as sentiment lexicons and part-of-speech taggers, are in their infancy with Farsi.

We are examining various machine learning methods, including meta-learning, to help improve the accuracy of sentiment classification for this problem.

Labeled sentiment of tweets in Farsi about the U.S.