Using word embeddings to probe sentiment associations of politically loaded terms in news and opinion articles from news media outlets

Abstract

This work describes an analysis of political associations in 27 million diachronic (1975–2019) news and opinion articles from 47 news media outlets popular in the United States. We use embedding models trained on individual outlets content to quantify outlet-specific latent associations between positive/negative sentiment words and terms loaded with political connotations such as those describing political orientation, party affiliation, names of influential politicians, and ideologically aligned public figures. We observe that both left- and right-leaning news media tend to associate positive sentiment words with terms used to refer to members of their own political in-group and negative sentiment words with terms used to denote members of their ideological outgroup. Outlets rated as centrist by humans display political associations that are often milder but similar in orientation to those of left-leaning news organizations. A weighted average of political associations by outlets’ readership volume hints that political associations embedded in left of center news outlets might have larger societal reach. A chronological analysis of political associations through time suggests that political sentiment polarization is increasing in both left- and right-leaning news media contents. Our approach for measuring sentiment associations of words denoting political orientation in outlet-specific embedding models correlates substantially with external human ratings of outlet ideological bias (r\u2009>\u20090.7). Yet, specific sentiment associations are sometimes multifaceted and challenging to interpret. Overall, our work signals the potential of machine learning models derived from news media language usage to quantify the ideological bias embedded in news outlet content.