This is the code for the final research project of Emory University CS329 Computational Linguistics
-
We applied NLP techniques including Linear discriminant analysis (LDA) with Gibbs sampling and Bert-based sentiment analysis on over 30,000 news articles from major US news media and conducted a topic-based analysis to detect their potential political leanings
-
A paper summarizing our work can be found here
Major studies in the field of news media analysis focus mostly on the categorization of media in terms of their political wings and little on differentiating their views on specific political topics. Our project seeks to examine media perspectives by using topic extraction with LDA to target more specific topics and conducting opinion mining via sentiment analysis. We attempt to present the public with a more authentic and clear view in regards to the media's political tendencies so that they can be more informed about their own bias toward major new outlets and also gain a more objective view regarding what to expect when reading news articles. Among the political topics examined, our study finds that the media generally hold negative views toward topics under coronavirus with different negativity, and have more various opinions toward topics under immigration but also choose different topics to cover. We also observe that different media hold notably more negativity toward some topics and also share their degree of negativity across different subtopics. Our result offers the public an intuitive way to understand media perspectives regarding specific political topics so that they may interpret information presented in news with a more objective attitude.
-
Yunjie(Ruby) Wu @yunjiewu777
-
Xinran(Alexandra) Li @shinrannli
-
Jingyu(Eula) Wang @EulaFRL