Analyzing Polarization in Social Media: Method and Application to Tweets on 21 Mass Shootings

“Train GloVe embeddings (Pennington et al., 2014) on V based on 11-50k random samples of tweets from each event.”

“com/ddemszky/framing-twitter 2016) and Facebook (Bakshy et al., 2015).”

“, 2009), and so emotional expression represents another semantic layer relevant to polarization (Iyengar et al., 2012; Suhay, 2015).”

“, forthcoming), and the propagation of partisan frames can influence public opinion (Chong and Druckman, 2007) and party identification (Fiorina and Abrams, 2008).”

“The strategy of highlighting particular aspects within topics as a means of framing (Entman, 2007) has also been quantified in the NLP literature (Boydstun et al.”

“We build on the ability of vector space models to represent higher-level semantics to develop our own embedding-based topic assignment approach, comparing it with two traditional LDAbased methods: MALLET8 and the Biterm Topic Model (BTM) (Yan et al., 2013); BTM was developed specifically for tweets.”

“, forthcoming), and the propagation of partisan frames can influence public opinion (Chong and Druckman, 2007) and party identification (Fiorina and Abrams, 2008).”

“Affect is intimately tied to ideological reasoning (Redlawsk, 2002; Taber et al., 2009), and so emotional expression represents another semantic layer relevant to polarization (Iyengar et al.”

“Modality is a lexical category concerned with necessity and possibility (Kratzer, 2002; Fintel, 2006).”

“, 2012) and in various other domains (Hirsh and Peterson, 2009; Yarkoni, 2010; Pennebaker, 2011).”

“…literature on emotion and political ideology: conservatives score higher than liberals on subjective measures of fear (e.g. Jost et al., 2017; Federico et al., 2009; Hibbing et al., 2014) and disgust sensitivity is also associated with political conservativism (e.g. Inbar et al., 2009, 2012).”

“To compare the partisanship of tokens at the event- and topic-level, we also z-score the log odds ratios (Monroe et al., 2008) within events and topics; the most partisan tokens are reported in Appendix F.”

“Affect is intimately tied to ideological reasoning (Redlawsk, 2002; Taber et al., 2009), and so emotional expression represents another semantic layer relevant to polarization (Iyengar et al., 13Note that these words in fact have the largest difference (negative and positive, respectively) if we…”

“(2014), which takes advantage of homophily in the following behavior of users on twitter (Halberstam and Knight, 2016).”

“Since word-affect associations are highly domain dependent, we tailored an existing affect lexicon, the NRC Emotion Lexicon (Mohammad and Turney, 2013), to our domain via label propagation (Hamilton et al., 2016).”

“Several works find that frames have changed over time and between such events (Muschert and Carr, 2006; Schildkraut and Muschert, 2014), and that frames influence opinions on gun policies (HaiderMarkel and Joslyn, 2001).”

“…literature on emotion and political ideology: conservatives score higher than liberals on subjective measures of fear (e.g. Jost et al., 2017; Federico et al., 2009; Hibbing et al., 2014) and disgust sensitivity is also associated with political conservativism (e.g. Inbar et al., 2009, 2012).”

“Framing and polarization in the context of mass shootings is well-studied, though much of the literature studies the role of media (Chyi and McCombs, 2004; Schildkraut and Elsass, 2016) and politicians (Johnson et al.”

“Prior NLP work in this area has considered how to extract factual information on gun violence from news (Pavlick et al., 2016) as well as quantify stance and public opinion on Twitter (Benton et al., 2016) and across the web (Ayers et al., 2016); here we advance NLP approaches to the public discourse surrounding gun v…”

“…accord with existing literature on emotion and political ideology: conservatives score higher than liberals on subjective measures of fear (e.g. Jost et al., 2017; Federico et al., 2009; Hibbing et al., 2014) and disgust sensitivity is also associated with political conservativism (e.g. Inbar…”

“, 2017) and that emotive political tweets are more likely to be shared (Stieglitz and Dang-Xuan, 2012).”

“, 2012) and in various other domains (Hirsh and Peterson, 2009; Yarkoni, 2010; Pennebaker, 2011).”

“Others have shown that emotion words can help detect political ideology on Twitter (Preoţiuc-Pietro et al., 2017) and that emotive political tweets are more likely to be shared (Stieglitz and Dang-Xuan, 2012).”

“The strategy of highlighting particular aspects within topics as a means of framing (Entman, 2007) has also been quantified in the NLP literature (Boydstun et al., 2013; Card et al., 2015; Naderi and Hirst, 2017).”

“C L ] 4 A pr 2 01 9 problems (Drakulich, 2015); in the mass shooting context, race is a factor in an event’s newsworthiness (Schildkraut et al., 2018) and is often mentioned prominently in media coverage, particularly when the shooter is non-white (Mingus and Zopf, 2010; Park et al., 2012).”

“, 2009), and so emotional expression represents another semantic layer relevant to polarization (Iyengar et al., 2012; Suhay, 2015).”

“Topic choice can be a tool for agenda-setting by establishing what an author or institution deems worthy of discussion (McCombs, 2002), and works in NLP have used topic modeling as an approach to measure this effect (Tsur et al., 2015; Field et al., 2018).”

“We estimate the party affiliation of users in the dataset from the political accounts they follow using a method similar to that of Volkova et al. (2014), which takes advantage of homophily in the following behavior of users on twitter (Halberstam and Knight, 2016).”

“These findings support related work (Perea, 1997; Delgado and Stefancic, 2017) discussing binary conceptualization of race in the US, and its influence in determining whether a shooter’s mental health or aspects of their identity are discussed.”

“Topic choice can be a tool for agenda-setting by establishing what an author or institution deems worthy of discussion (McCombs, 2002), and works in NLP have used topic modeling as an approach to measure this effect (Tsur et al., 2015; Field et al., 2018).”

“Several works find that frames have changed over time and between such events (Muschert and Carr, 2006; Schildkraut and Muschert, 2014), and that frames influence opinions on gun policies (HaiderMarkel and Joslyn, 2001).”

“…literature on emotion and political ideology: conservatives score higher than liberals on subjective measures of fear (e.g. Jost et al., 2017; Federico et al., 2009; Hibbing et al., 2014) and disgust sensitivity is also associated with political conservativism (e.g. Inbar et al., 2009, 2012).”

“Duxbury et al. (2018) find that media representations of white shooters disproportionately divert blame by framing them as mentally ill while representations of non-white shooters are more frequently criminalized, highlighting histories of violent behavior.”

“problems (Drakulich, 2015); in the mass shooting context, race is a factor in an event’s newsworthiness (Schildkraut et al.”

“C L ] 4 A pr 2 01 9 problems (Drakulich, 2015); in the mass shooting context, race is a factor in an event’s newsworthiness (Schildkraut et al., 2018) and is often mentioned prominently in media coverage, particularly when the shooter is non-white (Mingus and Zopf, 2010; Park et al., 2012).”

“The strategy of highlighting particular aspects within topics as a means of framing (Entman, 2007) has also been quantified in the NLP literature (Boydstun et al., 2013; Card et al., 2015; Naderi and Hirst, 2017).”

“Prior NLP work in this area has considered how to extract factual information on gun violence from news (Pavlick et al., 2016) as well as quantify stance and public opinion on Twitter (Benton et al., 2016) and across the web (Ayers et al., 2016); here we advance NLP approaches to the public…”

“, 2018) and is often mentioned prominently in media coverage, particularly when the shooter is non-white (Mingus and Zopf, 2010; Park et al., 2012).”

“, that polarized messages are more likely to be shared (Zafar et al., 2016) and that certain topics are more polarizing (Balasubramanyan et al.”

“Framing and polarization in the context of mass shootings is well-studied, though much of the literature studies the role of media (Chyi and McCombs, 2004; Schildkraut and Elsass, 2016) and politicians (Johnson et al., 2017).”

“Prior NLP work in this area has considered how to extract factual information on gun violence from news (Pavlick et al., 2016) as well as quantify stance and public opinion on Twitter (Benton et al., 2016) and across the web (Ayers et al., 2016); here we advance NLP approaches to the public discourse surrounding gun v…”

“We evaluate on the Indian English dialect feature detection task of Demszky et al. (2019) because this is one of the only available datasets to be handlabeled by a domain expert.”

“The work of [Demszky et al. 2019] analyzed the attitudes about 21 mass shooting event.”