Analyzing Mass Media influence using natural language processing and time series analysis

TL;DR

This study combines natural language processing and time series analysis to investigate how mass media influences public opinion, using the 2016 U.S. presidential campaign as a case study.

Contribution

It introduces a novel methodology integrating NLP and time series analysis to examine media influence on public opinion, emphasizing the combined role of topics and sentiment.

Findings

Sentiment alone does not explain poll variations.

Topics coverage combined with sentiment offers better insights.

Methodology is adaptable to other topics.

Abstract

A key question of collective social behavior is related to the influence of Mass Media on public opinion. Different approaches have been developed to address quantitatively this issue, ranging from field experiments to mathematical models. In this work we propose a combination of tools involving natural language processing and time series analysis. We compare selected features of mass media news articles with measurable manifestation of public opinion. We apply our analysis to news articles belonging to the 2016 U.S. presidential campaign. We compare variations in polls (as a proxy of public opinion) with changes in the connotation of the news (sentiment) or in the agenda (topics) of a selected group of media outlets. Our results suggest that the sentiment content by itself is not enough to understand the differences in polls, but the combination of topics coverage and sentiment content provides an useful insight of the context in which public opinion varies. The methodology employed in this work is far general and can be easily extended to other topics of interest.