As social news aggregator websites become more and more popular, researchers are inspired to harness the influx of information by classifying news articles without the need to read and analyze each one personally. Text analysis has been traditionally used to assign a classification to web articles, but suffer from low accuracy and the need to use training data. The ability to correctly assign a classification to a popular news article can mean the ability to generate more revenue by channeling the appropriate audience to said article.
University of Michigan researchers have developed a method to predict the political affiliation of users and news articles without the need to use training data as required by conventional supervised learning methods (SVM). They tested three different algorithms on field data obtained from Digg.com, a website where users can vote for the articles that they enjoy reading. They found that the best algorithm achieved a 96.3% accuracy on predicting the political leaning (conservative or liberal) of user and articles, as compared to a 92.0% accuracy from conventional SVM methods.
Applications and Advantages
- Online social news aggregators
- Search engine algorithms
- Classification of online blogs, articles, and users
- Generating ads that tailor to the majority of users who visit a certain site
- Higher accuracy
- Eliminates the need of training data and the costs of acquiring them