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.”

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

“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.”

“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).”

“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).”

“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).”

“…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.”

“Topic choice has been commonly used in NLP (Tsur et al., 2015; Field et al., 2018; Demszky et al., 2019) as a proxy for agenda-setting , the strategic highlighting of what aspects of a subject are worth discussing (McCombs, 2002).”

“The internet has often been accused of increasing political polarization among highly active users online (Demszky et al., 2019; Iandoli et al., 2021; Kubin & von Sikorski, 2021; Yarchi et al., 2021).”

“For finetuning on labelled data, we trained for three epochs with a batch size of 32, which corresponds to default settings recommended by Devlin et al. (2019).”

“In short, we calculated and visualized statewide average stance per drug per wave for geoinfer-ence; We follow Demszky et al 34 to infer political affiliation and Li et al 35 to infer healthcare background.”

“…as a middle ground between event-specific frames, emerging from the data and capturing properties unique to the given topic (Tsur et al., 2015; Demszky et al., 2019), and general issue frames (Boydstun et al., 2014; Card et al., 2015b; Johnson et al., 2017a; Field et al., 2018; Hartmann et…”

“…understand positions or discourses of hate based on dialectic confrontations, which currently represent a sphere which has still not been explored to any great extent in the literature focused on expressions of the identity of the young population in connection with gender in digital environments.”

“Unlike [12] that focused on a specific type of events (mass-shootings) and [15] that studied controversy surrounding the most-recent supreme court confirmation, we consider a longer, continuous time-horizon (2017 - 2020) within which two (or more) sub-communities discuss a broad range of issues.”

“…et al., 2020), ideology (Hirst et al., 2010; Iyyer et al., 2014; Preoţiuc-Pietro et al., 2017), and reasoning on certain topics (Egami et al., 2018; Demszky et al., 2019; Camp et al., 2021), to factual information, such as main topics (Gottipati et al., 2013), events (Trappl, 2006; Mitamura et…”

“Increased protests that resulted after the shooting at Marjory Stoneman Douglas High School in early 2018, along with evidence from increases in online discussions that follow school shootings (Demszky et al. 2019), suggests these eventsmay activate the knowledge, skills, and behaviors likely to promote voting.”

“1 3 of the crisis, there were quite a few [5, 96] that argued that the meeting of computer science and applied sciences is more than just adapting large data repositories and tools to analyze them.”

“While less-supervised approaches exist, they often use word statistics or hierarchical topic models, which lead to the same interpretability, stability, and simplification concerns as in §3.1 (Nguyen et al., 2013; Roberts et al., 2016; Demszky et al., 2019; Bhatia et al., 2021).”

“Many papers have recently been published on detecting political bias of online content either automatically (Baly et al. 2020; Demszky et al. 2019) or manually (Ganguly et al. 2020; Bozarth, Saraf, and Budak 2020).”

“Framing is used to study the political bias and polarization in social and news media (Tsur et al., 2015; Baumer et al., 2015; Card et al., 2015; Field et al., 2018; Demszky et al., 2019; Fan et al., 2019; Roy and Goldwasser, 2020).”

“Several works have looked at analyzing political ideologies, beyond the left and right divide, using text (Sim et al., 2013; Preo¸tiuc-Pietro et al., 2017), and specifically using Twitter data (Conover et al., 2011; Johnson and Goldwasser, 2016; Mohammad et al., 2016; Demszky et al., 2019).”

“The growing political polarization of the American electorate over the last several decades has been widely studied and documented [1, 2, 3, 4, 5, 6, 7, 8].”

“LIWC has been deployed to characterize psychological aspects related to political partisanship, such as affect states and pronouns in stances regarding mass shootings [Demszky et al. 2019], honesty and cognitive complexity in USA presidency candidates [Slatcher et al.”

“Given various conceptualizations of polarization, different frameworks have been developed to quantify pathological patterns of communication across groups holding similar or opposed stances on a given issue (Darwish 2019; Demszky et al. 2019; Morales et al. 2015; Weber et al. 2013).”

“Demszky et al. (2019) analyzed political polarization in social media and Voigt et al. (2017) examined the connections be-tween police officers’ politeness and race by analyzing language.”

“This body of work ranges from studies exploring the role of polarization in social media [11, 14, 15, 18, 20, 64, 66] to platform design aimed at attenuating polarization [5, 60, 61].”

“While social media platforms foster open communication and have the potential to offer more democratic information systems, they have simultaneously facilitated divisions in society by allowing the spread of polarizing and incendiary content [23,54].”

“Demszky et al. [2019] provide an NLP framework to quantify and analyze political polarization in 4.4M tweets on 21 mass shooting events.”

“Similarly, Demszky et al. (2019) measured polarization based on the content of tweets around issues of gun policies for a series of documented mass shooting events.”

“…et al. 2015; Garimella et al. 2018; Morales et al. 2019), ideological radicalization (Grover and Mark 2019; Crawford, Keen, and Suarez-Tangil 2021; Mamié, Ribeiro, and West 2021), and political framing (Demszky et al. 2019; Mendelsohn, Budak, and Jurgens 2021; Reiter-Haas, Kopeinik, and Lex 2021).”

“…Muise et al. 2022), polarization and segregation effects have been observed in diverging public narratives about society and signi fi cant events (Demszky et al. 2019; Li et al. 2017), online knowledge curation (Yang and Colavizza 2022), as well as behavior on social media (Adamic and Glance…”

“Demszky et al. (2019) highlight four linguistic dimensions of political polarization in social media: topic choice, framing, and affect an apparent force.”

“Analysis of religious polarization in our corpus (along the lines of political polarization in [13]) would require a reliable estimate of religious af-filiation.”

“From policy debates (Demszky et al., 2019) to modern conflicts (Palakodety et al.”

“Prior work has investigated the linguistic variation across different online communities, including linguistic differences between Republicans and Democrats (Demszky et al. 2019), politically homogeneous vs. heterogeneous groups (An et al. 2019), and overall stylistic variation among different communities on different…”

“6We experimented with the method of Demszky et al. (2019) to expand the lexicon for the domain by training GloVe embeddings on the dataset, and then for each emotion, retrieving the words that have the highest mean cosine similarity to the words associated with that emotion.”

“A large body of related work in NLP focuses on detecting stance, ideology, or political leaning (Baly et al., 2020; Bamman and Smith, 2015; Iyyer et al., 2014; Johnson et al., 2017; Preoţiuc-Pietro et al., 2017; Luo et al., 2020a; Stefanov et al., 2020).”

“There is extant discussion about the role of social media companies in fostering political polarization and how they should be regulated to avoid incentivizing extreme positions (Anderson and Huntington 2017; Demszky et al. 2019).”

“Clustering word embeddings has proven useful for discovering salient patterns in text corpora (Wilson et al., 2018; Demszky et al., 2019).”

“, 2017) as a starting heuristic, or evaluating a labeling approach by comparing geolocation tags attached to posts with historical voting patterns (Demszky et al., 2019).”

“The expression of emotional states constitutes a semantic level relevant to polarization and can assist in detecting ideological levels [6].”

“In NLP, a growing body of work has discovered mechanisms by which polarization manifests itself linguistically (e.g., Demszky et al., 2019).”

“, in [17], employ NLP techniques to uncover linguistic dimensions of political polarization.”

“On the other hand, stance detection has been used to analyze the attitude related to disruptive events, such as mass shooting event[5].”

“Of late, several works attempted to solve such tasks (Iyyer et al., 2016; Han et al., 2019; Demszky et al., 2019; Kulkarni et al., 2018; PreoţiucPietro et al., 2017; Diermeier et al., 2012).”

“Much attention has been paid to opinion polarization in online settings, such as social media [9], [23], [26]–[30], among political blogs [31], [32], and in comment sections [15].”

“Demszky et al. (2019) apply ABSA methods to pioneer tweets on the US mass shootings topic, where the topic was politically discussed from different viewpoints according to the locations of events with the contrasting use of the terms “terrorist” and “crazy”, that contribute to polarization.”

“In order to identify topic areas within which to measure misalignment, we implement a topic assignment procedure, inspired by the success of a simple embedding-based approach for Twitter data in Demszky et al. (2019).”

“Prior research in the area has found that online groups can tend to develop their own discourses and vocabulary around certain topics or events (Darwish et al. 2018; Demszky et al. 2019; Graells-Garrido, Baeza-Yates, and Lalmas 2020; Li, Porco, and Goldwasser 2018; Stefanov et al. 2020).”

“, 2021), or by using data-driven methods (Demszky et al., 2019; Roy and Goldwasser, 2021).”

“discussions of long-discussed and highly politicized topics such as gun control [25] and presidential elections [26].”

“For this purpose, we construct positive and negative sentiment dimensions in our embedding space using our centroid method, with positive and negative seed lexicons taken from Demszky et al. (2019) (see Appendix A.4 for details).”

“These groups can be either pre-defined , through the use of manual labeling [2, 3, 5, 33, 40, 53, 65] or automated labeling [24, 32, 84], or partitioned using unsupervised methods such as community detection [22, 31, 37, 54, 63].”

“Different from (Demszky et al., 2019) that assigns only one topic with the highest probability to a document, we allow a document to be assigned multiple topics with different probabilities.”

“Such techniques have been used in various disciplines such as [6] in Political Science and [7] in mental health realm.”

“…then obtained the comments from the unlabeled corpus whose embeddings were closest to the seed embeddings (i.e. the nearest neighbors). following [44, 43] cosine distance was used as the distance metric. can someone tell me where i can help charity to countries should take a stand for these…”

“Demszky et al. [33] presented a study of 21 U.S. mass shooting events to measure polarization in common frames in X .”

“Demszky et al. (2019) investigated the usage of expressions for necessity modality (including should, must, have to and need to) among tweets about mass shooting events, as necessity modality bears the illocutionary force of calling for action or change in the discourse under discussion.”

“Demszky et al. (2019) looked at 4.4 million tweets from 21 mass shootings.”

“Despite the early euphoria about social media’s potential for bringing global democratization [3] and allowing cross-cutting interactions to promote the exchange of different political views [4], researchers are increasingly concerned about social media’s role in facilitating polarization [5].”

“e annotated based on authorship. We envision an approach where we transfer knowledge from tweets to long text at test time. While previous work has attempted to analyze tweets for political sentiment [3], there is no research on domain adaptation from short to long documents in this context. There has been research on…”

“The effects of mass shootings on public discourse are well studied [27,66,67], from the general public’s interest [25] to media coverage [26], but much less is known about the response of policymakers to these tragic events.”

“be created using a variety of methods, and salient topics can be derived from tweets collected using both traditional LDA and non-traditional methods (Demszky et al., 2019).”

“Demszky et al. (2019) and Ziems and Yang (2021) inspect the usage of deon-2 tic modal verbs (e.g., ‘should’, ‘need’) in calling for actions, assigning blames and making moral arguments.”

“…scaffolding collapse , Pittsburgh bridge collapse , Fern Hollow Bridge Collapse , I-85 Overpass collapse } Keyword-based filtering is a well-established, high-recall approach to pruning datasets tailored for specific interests Demszky et al. (2019); Halterman et al. (2021); Dutta et al. (2022).”

“While the polarization literature has examined the context of mass shootings (Demszky et al., 2019) and vigilantism (Kurumathur et al., 2022), little attention has been given to understanding the public’s support for or opposition to vigilantism.”

“I modify the topic model introduced in (Demszky et al. 2019), which combines sampling and vector spaces to encourage topics that are independent of subreddit size or focus.”

“Both political polarization and hate speech are mainly manifested through language that activates certain frames (Demszky et al., 2019).”

“Algorithms have been used to diagnose broader cultural biases and stereotypes, often through use of word embeddings [11, 14, 25, 29]; this work differs from ours because it studies”