Three Facets of Online Political Networks: Communities, Antagonisms and Controversial Issues

Abstract

Without any doubt, for the last decade, online social networks have been hubs for political participation both for elite members such as politicians, parliamentary members, and ordinary citizens. Each online activity leaves digital traces of the users participating which enable researchers to study online political behavior and build multi-faceted sensors to understand the underlying dynamics. In this talk, I broadly focus on three facets of online political behaviors of users and how to model them. First, I investigate the formations of communities in online political networks. Second, I focus on detecting negative linkages between users such as enmities or antagonisms. Œird, I present the problem of the controversial issue and frame detection in online political networks. For each part, I give one exemplary recent work and possible future directions to take.

BACKGROUND AND SUMMARY

Online social networks have been aŠracting people from a wide range of political spectrum for the last decade to express their political opinions, share the news that they care with their contacts, form groups and debate with or against each other. Beyond individual participation, protests are organized [ 7 ], [ 12 ], new alt-right movements emerge [ 3 ], terrorist organizations recruit [ 11 ] and many other political activities reside in online environments. Most importantly they leave an immense amount of digital footprints while participating. Œese traces of people’s online political activities give researchers unequivocal opportunities to analyze, reason and predict the human behavior in online political seŠings.

Œe challenges of conventional social media mining research is present for political sub-domain of social media data as well. Sparsity of the user network and short and noisy nature of textual content constitute the major body of the challenges. Moreover, in the political sub-domain a high level of irony and indirectness are involved. In this talk, I focus on three dimensions of multi-faceted human activity in online political networks, online communities and what do users have in common in a politically pure online community, antagonisms, enmities or any other form of negative linkages between users in online political networks, and controversial issues causing polarization in online political networks and frames each side puts forward for issues.

Further details for each dimension are given in Section 1.1, 1.2, and 1.3, respectively. I give one recent proposed model for each dimension of online political activities of users and possible future directions for research. 1.1

Community Detection

Community detection is an integral part of studying online political networks. A community in an online political network is a group of users which interact with each other more than they interact with others and share similar characteristics in terms of their political belongings. Many studies show that endorsement network of users gives invaluable piece of information to detect underlying politically pure communities. For example, in the political sub-domain of TwiŠer, retweet network of politically motivated users uncover the political identities of groups [ 1 ], broadly. However, the sparse nature of endorsement networks challenges to identify political groups in complete chunks. Many users do not endorse each other although they propagate similar political contents. It may be simply because they do not know each other.

In this talk, a recent work [ 8 ] that tackles the sparsity problem by employing network-independent features of users is presented. Œe role textual content, shared article URLs and platform speci€c content tags (i.e. hashtags) play in formation of communities is studied by proposing three di‚erent non-negative matrix factorization frameworks. Œe work also incorporates social balance theory to introduce new arti€cial endorsement links between users to overcome the sparsity problem. Possible future directions are investigated in three folds; temporal dynamics of communities, emerging and fading communities. of it. Œe more distant the vectors of the issue are the more di‚erent frames are used for the issue by each communities.

1.2 Negative Link Prediction

Capturing disagreements, antagonisms, and enmities constitutes another vital dimension of understanding online political activities of users. Œey may be even more informative than detecting alliances, agreements or coalitions in the seŠings where endorsement and sharing is motivated by the online platform. Two typical examples are Facebook and TwiŠer in which features of endorsement and sharing is available in di‚erent forms such as retweet, like or share while opposition, disagreement or any form of denouncement are not. Features of platform-speci€c positive interactions help us to understand the nature of positive links between users. However, capturing negative links is a much more challenging task since there is no explicit “dislike” feature in major online social network platforms.

In the talk, a recent work [ 10 ] which suggests exploring sentiment signals in textual interactions, platform-speci€c positive interactions and their composition for negative link prediction is brieƒy discussed. Œe work assumes that if textual interactions of two users (i.e. mentioning in TwiŠer) carry many negative sentiment words, two users likely to form negative link between each other. It further assumes two users to have a positive link in between if they use any platform-speci€c positive interactions towards each other. It also follows simple social balance rules such that enemy of my enemy is my friend and friend of my friend is my friend to advance the link prediction model. In the talk, two applications of the framework are presented to show the added value of the negative links in detecting communities more accurately and in describing rivalries and coalitions more e‚ectively. 1.3 Issue and Frame Detection Œe last facet of understanding online political activities of users this talk focuses is detecting controversial issues and frames. Online social network users express their opinions on various political issues. Needless to say, each user brings its own perspective and nuances are inevitable. Characterizing which issues polarize online crowds, while which issues bring communities closer is a key to understand the dynamics behind formation of coalitions and conƒicts. Moreover, capturing the sides communities take for a controversial issue correctly is also valuable. Although it is a well studied problem for conventional text documents such as parliamentary ƒoor debate records, a few e‚ort has been put forward for detecting controversial issues for online social networks [ 4 ] [ 5 ] [ 6 ].

By [ 2 ], framing is de€ned as selecting some aspects of a perceived reality and make them more salient in a communicating text. A recent work [ 9 ] that incorporates the concept of framing to capture controversial issues into a word embedding model is discussed to illustrate a solution to the problem of issue and frame detection. It proposes to model dimensions of embedding space as frames that are highlighted by di‚erent groups. Œen, it computes a unique community speci€c vector representation of each issue. Œe distance between community speci€c word vector representations of each issue becomes an e‚ective metric to quantify the controversy

References

[1] Michael Conover , Jacob Ratkiewicz, MaŠhew R Francisco, Bruno Gon¸calves, Filippo Menczer, and Alessandro Flammini . 2011 . Political Polarization on TwiŠer . ICWSM 133 ( 2011 ), 89 - 96 .

[2] Robert M. Entman . 1993 . Framing: Toward Clari€cation of a Fractured Paradigm . Journal of Communication 43 , 4 ( 1993 ), 51 - 58 .

[3] Gabriel Emile Hine , Jeremiah Onaolapo, Emiliano De Cristofaro, Nicolas Kourtellis, Ilias Leontiadis, Riginos Samaras, Gianluca Stringhini, and Jeremy Blackburn . 2017 . Kek, Cucks, and God Emperor Trump: A Measurement Study of 4chan's Politically Incorrect Forum and Its E‚ects on the Web .

[4] Kristen Johnson and Dan Goldwasser . 2016 . ”All I know about politics is what I read in TwiŠer”: Weakly Supervised Models for Extracting Politicians' Stances From TwiŠer . In COLING.

[5] Kristen Johnson and Dan Goldwasser . 2016 . Identifying Stance by Analyzing Political Discourse on TwiŠer . NLP+CSS 2016 ( 2016 ), 66 .

[6] Aditya Joshi , Pushpak BhaŠacharyya, and Mark James Carman. 2016 . Political Issue Extraction Model: A Novel Hierarchical Topic Model Œat Uses Tweets By Political And Non-Political Authors . In Proceedings of the 7th WASSA@NAACLHLT 2016 , San Diego, California, USA. 82 - 90 .

[7] Josep-Llu ´ ıs Mico´ and Andreu Casero-Ripolle´s. 2014 . Political activism online: organization and media relations in the case of 15M in Spain . Information, Communication & Society 17 , 7 ( 2014 ), 858 - 871 .

[8] Mert Ozer , Nyunsu Kim, and Hasan Davulcu . 2016 . Community detection in political TwiŠer networks using Nonnegative Matrix Factorization methods . In 2016 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM) . 81 - 88 .

[9] Mert Ozer , Amin Salehi, and Hasan Davulcu . 2017 . Detecting Controversial Issues and Related Frames in Polarized Online Political Networks . In NLP+CSS 2017 . submiŠed.

[10] Mert Ozer , Mehmet Yigit Yildirim, and Hasan Davulcu . 2017 . Negative Link Prediction and Its Applications in Online Political Networks . In Še 28th ACM Conference on Hypertext and Social Media (ACM Hypertext 2017 ). to appear.

[11] Andrew Stanton , Amanda Œart, Ashish Jain, Priyank Vyas, Arpan ChaŠerjee, and Paulo Shakarian . 2015 . Mining for causal relationships: A data-driven study of the islamic state . Vol. 2015-August. Association for Computing Machinery , 2137 - 2146 .

[12] Zeynep Tufekci and Christopher Wilson . 2012 . Social media and the decision to participate in political protest: Observations from Tahrir Square . Journal of Communication 62 , 2 ( 2012 ), 363 - 379 .