Good News AI Investigating feasibility of categorizing positive sentiment in general news

Abstract

In today’s society we are constantly fed information about catastrophic or sad events through media. While it is important to know about these events, it should be equally important to also see all the good things that are happening in our world. Therefore, this thesis proposes two algorithms for classifying full-length news articles to remove the non-positive articles. Traditionally these types of algorithms require a large amount of labelled data, but this thesis explores possibilities for sentiment classification with a limited amount of labelled data. The best performing algorithm presented is this thesis achieves a precision percentage of 98% with only 40 articles used for training.