Finding Topics Associations in Political News Using Deep Learning

• In Event: Computational Methods in Journalism Research

Abstract

Associations between topics are a key feature in the process of understanding the role of topics in texts, as they may be indicative for issue framing or the understanding of meaning making in content and context of messages. Analyzing the association requires a preliminary analysis of the corpus using a topics classification method for text analysis, but current methods are either expensive or less appropriate to hypothesis testing. In this paper we present a process that combines topic modeling and a deep neural network model that together create a method for topic classification that faces this challenge. In the paper we show the efficiency and usefulness of the method, explain how to set up and validate the method, and elaborate the different layers of the deep model. We demonstrate this process on news article collected from news websites, showing that it is resulting in a meaningful identification of issue frames.

Authors

• Yair Fogel-Dror, The Hebrew U of Jerusalem

• Shaul Shenhav, Hebrew U of Jerusalem

• Tamir Sheafer, Hebrew U