Finding a News Article Related to Posts in Social Media: The Need to Consider Emotion as a Feature

Abstract

As social media data grows to a tremendous size, understanding posts in social media becomes important for many applications such as commercial or political analysis. It is helpful because it gives us insight into how significant social issues affect a person or a group of people. Therefore, this paper proposes a method to find a news article that relates to a user's posts on Facebook. A classification model based only on keywords does not work well because there are different news articles with similar keywords. We propose adding an emotion feature to the classification model to handle this problem, as we observed that many news articles have a distinguishing emotional distribution. We show that classification models with an emotion feature yield better performance than models without an emotion feature. Furthermore, a classification model with an emotion feature works well when there is apparent emotion, and it does not perform well if there is language play or puns in the text.