Justin H. Gross

The Media Frames Corpus: Annotations of Frames Across Issues

We describe the first version of the Media Frames Corpus: several thousand news articles on three policy issues, annotated in terms of media framing. We motivate framing as a phenomenon of study for computational linguistics and describe our annotation process.

Measurement and theory in legislative networks: The evolving topology of Congressional collaboration

Abstract The examination of legislatures as social networks represents a growing area of legislative scholarship. We examine existing treatments of cosponsorship data as constituting legislative networks, with measures aggregated over entire legislative sessions. We point out ways in which the direct application of models from the social networks literature legislative networks aggregated over entire sessions could potentially obscure interesting variation at different levels of measurement. We then present an illustration of an alternative approach, in which we analyze disaggregated, dynamic networks and utilize multiple measures to guard against overly measure-dependent inferences. Our results indicate that the cosponsorship network is a highly responsive network subject to external institutional pressures that more aggregated analyses would overlook.

Twitter Taunts and Tirades: Negative Campaigning in the Age of Trump

What drives candidates to “go negative” and against which opponents? Using a unique dataset consisting of all inter-candidate tweets by the 17 Republican presidential candidates in the 2016 primaries, we assess predictors of negative affect online. Twitter is a free platform, and candidates therefore face no resource limitations when using it; this makes Twitter a wellspring of information about campaign messaging, given a level playing-field. Moreover, Twitter’s 140-character limit acts as a liberating constraint, leading candidates to issue sound bites ready for potential distribution not only online, but also through conventional media, as tweets become news. We find tweet negativity and overall rate of tweeting increases as the campaign season progresses. Unsurprisingly, the front-runner and eventual nominee, Donald Trump, sends and receives the most negative tweets and is more likely than his opponents to strike out against even those opponents who are polling poorly. However, candidates overwhelmingly “punch upwards” against those ahead of them in the polls, and this pattern goes beyond attacks against those near the top. Sixty of 136 dyads are characterized by lopsided negativity in one direction and only one of these 60 involves a clearly higher status candidate on the offensive.

Analyzing Framing through the Casts of Characters in the News

We present an unsupervised model for the discovery and clustering of latent “personas” (characterizations of entities). Our model simultaneously clusters documents featuring similar collections of personas. We evaluate this model on a collection of news articles about immigration, showing that personas help predict the coarse-grained framing annotations in the Media Frames Corpus. We also introduce automated model selection as a fair and robust form of feature evaluation.