Finding Political Blogs and Their Political Leanings

The Blogosphere has more influence on both the general public’s opinions and mainstream media nowadays. As vast political blogs representing grassroots voices go online, it is both interesting and useful to find them and determine their political leanings as either liberal or conservative. In this paper, we address both problems by using front pages of blogs in a corpus we built to create two classifiers: one to classify arbitrary blogs as political or non-political; the other to classify political blogs as liberal or conservative. We explored performance by using ‘bag of words’ and link features with the help of the SVM classification algorithm. Results show that a political vs. non-political blog classifier can achieve an overall accuracy of 91.40% by merely using ‘bag of words’ features. Combining both ‘bag of words’ and some link features, the best liberal vs. conservative blog classifier can on average predict a political blog’s political leaning with an accuracy of 86.63%. With the help of some feature selection methods, however, analysis also indicates that potential in using simple ‘bag of words’ features in political leaning categorization is limited in that it lacks enough discerning power in detecting differences of opinions expressed by liberal bloggers from conservative ones.