Towards Conflictual Narrative Mechanics Laura Spillner1 , Carlo R.M.A. Santagiustina2,3 , Thomas Mildner1 and Robert Porzel1 1 University of Bremen, Bremen, Germany 2 Ca’Foscari University, Venice, Italy 3 Venice International University, Venice, Italy Abstract We propose a five steps methodology to retrieve, reconstruct and analyse conflict related narratives in a standardized and automated way. Our methodology combines AI and network analysis techniques to build a visual representation of key agents and entities involved in a conflict and to characterize their relations. Unlike the majority of existing methods, ours can be applied to any type of conflict, as, through two data downloading phases, it first generates a bird’s-eye representation and then a fine-grained map of any conflict. Given the broad applicability of the proposed methodology, we believe that this work moves the first steps towards a better understanding of conflictual narrative mechanics. 1. Introduction a concerning problem for users as well as public discourse to experience alternative points of views. Online social media have started as tools for people In a recent crisis, this problematic tendency be- to connect with others all over the globe [1, 2, 3]. came blatantly real when Russia started its invasive Meanwhile, platforms like Facebook, Reddit, and war against Ukraine on February 14th, 2022, es- Twitter have been labeled ‘game-changers’ for en- calating a fiery situation that began with Russia’s tertainment [1, 4] while offering novel opportunities annexation of the Ukrainian peninsula Crimea in to advertisers and businesses [5, 6, 7]. Increasing 2014. Soon after the war’s outbreak, on the 28th of the impact of social media even further, platforms February, 2022, the social network Twitter decided take an important part of the everyday political to expand labeling policies for content as a counter discourse [8, 9, 10]. Twitter in particular is used to the spread of misinformation on behalf of Russia, not only by people who share their opinions and adding the ‘Russia state - affiliated media’ tag to engage in political discussions, the platform is also related posts. Showing an impact of the spread used by official agencies as well as people working of Russia’s media outlets [11], Twitter’s decision in governments to disseminate information more moved Russia to block the platform in its nation, rapidly than via traditional news media. as well as Facebook who followed a similar strategy. When it comes to political and social conflicts, Both interventions, of Twitter and of Russia, high- the surrounding online conversation is often char- light how vulnerable public speech and information acterized by a range of opinions. Discourse may can become. In this particular case, Russia’s gov- evolve around conflict, parties involved therein, or ernment decided to cut off their country from major actions undertaken while different ‘causal’ relation- western media outlets, allowing them to precisely ships between events and actors may be asserted control news available to Russian’s citizens. in naïve manners. By being constantly exposed to Resulting conflictual narratives, occurring during these conversational dynamics, users’ opinions may crises like the one mentioned above, present an ur- be influenced by the narrative that is most popular gent need to understand underlying mechanics. In based on general interest of their personal social this paper, we present a methodology, as a work in media bubble. This effect is further amplified by progress, to study and investigate the online narra- tailored algorithms that elevate content predicted tives surrounding conflicts and crisis. The methodol- to be aligned with prior interest. This results in ogy itself is not necessarily limited to conflict alone, but aims to discover different perspectives on social IJCAI 2022: Workshop on semantic techniques for narrative-based understanding, July 24, 2022, Vienna, Aus- media while limiting any introduced researcher bias tria when constructing the corpus itself. This paper $ laura.spillner@uni-bremen.de (L. Spillner); describes the idea behind the methodology. We aim carlo.santagiustina@unive.it (C. R.M.A. Santagiustina); to use the methodology to analyse Twitter conversa- mildner@uni-bremen.de (T. Mildner); tions of Russia’s invasive war against Ukraine, and porzel@uni-bremen.de (R. Porzel) © 2022 Copyright for this paper by its authors. Use permitted under we present here the results of the first (preliminary) Creative Commons License Attribution 4.0 International (CC BY 4.0). part of this analysis. CEUR Workshop Proceedings http://ceur-ws.org ISSN 1613-0073 CEUR Workshop Proceedings (CEUR-WS.org) 19 Our hypothesis is that conflict narratives are 2.1. Conflict narratives strategically designed around recurrent story-telling Conflicts and their escalation in the physical and patterns and frames that assign a set of (asym- narrative space are generally the by-product of het- metric and stereotyped) roles to involved parties. erogeneous beliefs [12], asymmetric information [13], We plan to ultimately utilize the methodology pre- cognitive biases [14], like the availability heuristic sented herein to investigate this hypothesis on the and the confirmation bias, as well as complex entan- case-study of the Russo-Ukraine war; the results of glements [15, 16] of competing interests, strategies, this investigation will be presented in future work. and objectives, which are often opaque being diffi- When trying to capture the context and view- cult to elicit and model. point of narratives, one of the main difficulties is Conflict narratives, and more in general debates not to introduce researcher bias to sampled data: about conflicts through which these narratives By collecting, for example, only posts written in evolve and spread, have long been studied in the English, by defining which terms to query for, which social and political sciences under many different conflict-related narrative frames are being looked approaches and perspectives [17, 18, 19]. For ex- up, or which actors in the conflict are of interest ample, the study of conflict framing [20], factional apart from the main parties, any resulting corpus discourse design [21], and other polarizing commu- represents only a certain fraction of the discourse. nication strategies [22] is key for understanding any Therefore, we define a two-phase approach to consensus-building and group-mobilization process investigate conflict narratives based on online con- when there are competing views or interests at stake. versations on Twitter: The narrative dimensions of conflicts are gen- 1. A dataset is collected of tweets that explic- erally more visible in non-authoritarian countries, itly mention the conflict at hand through where partisan narratives [23] are constructed and the names of the main parties or neutral employed in relation to public support-building and conflict-related terms. From this, the enti- policy justification. Conflict narratives operate at ties involved in the conflict, such as actors, all levels of national and international governance events, and locations, either actively or pas- processes. Especially democratic countries aim to sively, are discovered and the most important mitigate and resolve potential or actual conflicts entities and their co-occurrence are identi- through transparent, informed, and participatory fied. deliberative processes. From local debates about 2. More fine-grained data is then collected by public policies to be implemented in response to searching for all the main entities and cap- a pandemic, like COVID-19, to the renegotiation turing frames used to characterize these enti- of trade agreements between parties that compete ties as well as possible relationships between for the control on strategic natural resources and them. technologies, like conflicts for rare earths or for the control over semi-conductor industry technologies, So far, we have conducted the first phase of the the emergence of conflict narratives is an ubiquitous analysis on the conflict between Russia and Ukraine, phenomenon in contemporary times. This is partic- as characterized by a sample of the Twitter conver- ularly evident in decentralized and multi-directional sation in the spring of 2022. In order to explain the online communication mediums such as social media proposed methodology, we present the results of this [24, 25], which have become the default propagation first step, detail the steps planned for the second medium for (popular) narratives, including conflict- phase analysis, and discuss which challenges remain related ones. to be solved in order to expand the methodology to a greater scope and other case studies. 2.2. Conflict modeling Recent attempts to model conflicts, like [26], have 2. Related Work shown that conflicts are not necessarily the outcome of diverging material interests among individuals The following section presents a brief summary of and groups, as modeled in early game-theoretic recent literature on conflict-related narratives with works [27]. They can also be the product of: (i) a particular focus on how these narratives are con- differently-biased or competing world views used to structed and diffused through online social media. decipher events and to comprehend the intentions underlying specific actions or communications by other individuals and groups [28]; or (ii) to the de- 20 terioration of inter-group trust [29]. As a result, interest that are involved in the conflict dynami- conflicts may emerge and exacerbate even when the cally from the online conversation surrounding it, material interests of the different parties converge instead of defining them a-priori. Thus, we hope to from a rational (i.e., utilitarian) perspective. For reduce the bias that would otherwise be introduced this reason, the presence of extrinsic incentives [30]during the corpus construction. Secondly, this also may not suffice to mitigate or resolve ongoing con- makes it possible to employ the same methodology flicts grounded on incompatible belief-systems or on to investigate extremely different topics and types the lack of trust, like during the cold-war [31]. For of conflicts. example, this is the case in attrition wars, which Although in this paper, we focus on the Russian may be represented as negative-sum games. Rather war against Ukraine, the aim for this approach is then immediately negotiating a mutually advanta- to be independent of any specific topic in order to geous agreement, two parties are ready to bear the enable other works to also re-construct narratives material and humanitarian costs of a long-lasting from different crisis, whether violent or non-violent, conflict if that gives them the opportunity to pun- where the parties involved are not clearly defined, ish the other, by reciprocating harm and causing a may change over time, or where the definition of similar or larger damage to their opponents. parties depends on specific viewpoints. In the following sub-sections, we present how 2.3. Conflict-related narratives extraction we are currently using this two phase approach to identify which narratives are shaping the online and events re-construction conversation on Titter concerning the ongoing war The incessant growth of online social media and in Ukraine. Phase 1 describes the analysis that has their communities [32], together with the increasing been conducted so far as well as the results of this, availability of computational power and advanced while Phase 2 discusses the steps that we plan to linguistic analysis methods [33, 34, 35] for big tex- conduct next. tual datasets offers an opportunity to capture and model conflict narrative dynamics on an unprece- 3.1. Phase 1 - Discovery dented scale. Recent works [36, 37, 38] have used posting ac- In this first discovery phase, the goal is to collect tivities on social media and online newspaper arti- tweets from a specified time interval which explicitly cles, including comment sections, for capturing and mention the conflict in question. We thereby collect mapping partisan or faction-specific arguments and any tweet that meets two conditions: Tweets that (1) their dynamics across time through a combination either contain the name of at least one of the directly of AI and network analysis methods. A further step involved parties or the name of a specific event and towards the automated mapping of conflicts and that (2) contain a generic term like conflict, tension, of their key actors and events has been done by or crises, denoting that the tweet refers to a conflict [39] and [40] who combined large-scale knowledge related to the selected event or involved parties. graphs with semi-structured sources in an event KG From these tweets, we identify all actors and entities RDF-representation. involved in the conflict which are mentioned most The former branches of research open the way to often in this context. We are then able to construct a new AI-augmented research field on all kinds of a network based on their co-occurences. conflicts. Such research could combine the poten- tials of NLP, network analysis, and computational 3.1.1. Step 1: Corpus construction linguistic methods with the semantic web, serving To construct the first corpus on the current war in as interfaces for the real-time observation and un- Ukraine, we collected all tweets derstanding of conflict-related narratives. • between the first of January and the first of 3. Methodology May 2022 • which contain the words: “Ukraine”, “Rus- To analyze the narratives that surround certain con- sia”, and any of “conflict(s)”, “tension(s)”, flicts, we developed a methodology consisting of “crises” or “crisis” two phases which include two cycles of data collec- • which are not retweets. tion. The reasons behind this are twofold: Firstly, it allows the discovering of actors and entities of As of now, we have only conducted a preliminary trial in order to test and validate our methodology. 21 For this, we have only searched for these terms in same holds for other entities involved, e.g. a person English. Hence, the collected corpus only represents being referred to as a “president” or “dictator”. Ei- a perspective from English-speaking Twitter users. ther variant will be collected if it appears in a noun However, we want to also collect tweets contain- phrase together with one of the named entities. ing these key terms in other languages, including In this step, we use the open source NLP library Ukrainian and Russian, in order to generate a mul- spaCy for part-of-speech tagging, dependency pars- tilingual corpus that represents a broader view and ing, and named entity recognition. From this, we contains differing narratives that include those from identify all noun phrases which include a token that both of the opposing parties. was classified as a named entity. Such identified en- An advantage of this two phase approach is that tities would include only geopolitical entities such as at the beginning, only a small number of terms needs countries, nationalities or religious/political groups, to be defined, which are compatible with any type organizations, persons, events, and locations. How- of conflict. We plan to employ the help of native ever, these entities would exclude instances such as speakers of other languages to translate these terms cardinals. and validate the results of the first phase. However, Identified noun phrases can then be consolidated we hope that we will not require the help of experts in entities, by linking them to named entities in a in the domain which are also native speakers of known knowledge graph such as WikiData (consoli- the additional languages. Apart from the definition dating e.g. “the dictator Putin”, “Vladimir Putin” of this small set of neutral conflict-related starting and “the russian president”, etc.). For the prelim- terms, the collection of the data (across both phases) inary analysis that follows, the consolidation step is entirely automatic. We chose to go with only these has yet to be implemented. In future versions of this three terms (conflict, tension and crisis) firstly in work this will be done through name matching and order to limit the scope of the twitter query, and name substring matching, without disambiguation, secondly because we believe that other synonyms or through more advanced distributional semantics of those terms generally connote either violent or and ML methods, like OpenTapioca [41]. non-violent conflicts (e.g. synonyms of “conflict” on ConceptNet include terms such as “battle” or 3.1.3. Step 3: Co-occurence network “disagreement”). For this preliminary analysis, we collected a total In the third step, identified entities are mapped of 724400 tweets. In a first pre-processing step, through an undirected network based on their co- we removed special UTF-8 characters, like emojis, occurences in the corpus. Each node represents a emoticons, and URL links. entity, with edge weights denoting how often two en- tities occur together in one tweet in the dataset. By so doing we obtain a weighted and undirected net- 3.1.2. Step 2: Entity Recognition work containing 933081 nodes and 10871807 edges. After generating this dataset, we use this broad- With an average degree of nodes equal to 23.303 coverage overview of the conflict at hand to discover and an average weighted degree of nodes equal to from it which actors or other entities are relevant to 46.162. the conflict. These will constitute the terms which Network nodes are then filtered based on their we will explicitly teach for in the second phase. Of weighted degree centrality metric; alternative cen- interest are trality measures, like pagerank, betweenness or eigen-centrality may also be used for this purpose. • named entities such as known persons, orga- This filtration is done in order to remove entities nizations, countries or peoples, e.g. in this that are less influential (more peripheral) and have case “Putin”, “Ukraine”, “EU”, “Russians”. a marginal role in the entity network for the se- • noun phrases which include these same terms lected conflict. As shown in Figure 2, this filtration as signifiers, such as “Ukrainian president”, step removes those entities which are mentioned “russian army”, “Ukraine war”, “russia- rarely in the dataset, like entities weekly related ukraine conflict”, etc. to the conflict and other irrelevant noun phrases. This might include entities such as sports teams or Therefore, instead of defining a term for the con- their fans who might be involved in a metaphorical flict and thus characterizing it ourselves – based on conflict, misspellings, or other noun phrases that the different connotations of “war”, “unrest”, “in- do not play a relevant role in online conflict-related vasion”, “operation”, etc. – we are able to discover narratives. which terms are used in online conversations. The 22 themselves, as well as the relationships between them, are characterized by different people online, for example through adjectives and verbs qualifying the relation between two key entities. Our hypothesis is that the narratives surround- ing the actors involved in the conflict are based on recurring phrase fragment patterns and frames that assign specific roles and attributes to the involved parties. In many cases, these roles are a-symmetric and mirror certain stereotypes that are common to (almost) all conflict narratives, like the role of the victim and that of the perpetrator. We plan to use the methodology we describe herein to investigate this hypothesis on the corpus we are currently col- lecting about the war in Ukraine. Using the tweets from this Phase 2 corpus, we aim to collect and assign a number of different frames to the actors and relations, which are essential constituents of the conflict narrative. For example, we expect that Figure 1: Entities network for the Ukraine-Russia con- we will find flict. Network filtered by node weighted degree: 0.999 percentile; and then by edge weight: 0.99 percentile. • verbs related to asymmetric roles in the con- flict, like: aggression / protection, offense / defense, attack / counter-attack, owner- 3.2. Phase 2 - Analysis of Narratives ship claims, “deserviness” claims, resisting/ surrendering, etc. In the second phase, we plan to retrieve thinner • conflict related nouns and adjectives, like: grained data about the conflict on the selected so- aggressor / aggressed, invader / invaded, lib- cial media, in order to (re-)construct and analyze erator / liberated, oppressor / oppressed, the narratives surrounding specific relational blocks. strong / weak, winning / losing, perpetrator As a starting point for this will serve the key ac- / victim, etc. tors/entities and relations (dyads of actors), which were identified in Phase 1 based on the chosen cen- • characterizations of the conflict or its escala- trality metric, together with the largest/heaviest tion, like: justified / unjustified, legitimate cliques of order 3+. / illegitimate, necessary /unnecessary, ex- plainable / unexplainable, expected / un- expected, hot / cold, violent / non-violent, 3.2.1. Step 4: Second corpus construction verbal /physical, ideological, political, eco- At this point, we have identified a number of key nomic, financial, military, etc. entities and relations, as well as different reference • equivalently, characterizations of terms which are used to refer to them. These will – the peoples or populations involved in then be used to construct a new set of queries from the conflict which a second corpus of tweets is collected. In this – the leaders of the factions involved in second phase, we are going to search explicitly for the conflict the entities that the first phase identified as being perceived as important actors in the conflict, using – the countries involved in the conflict the terms that were discovered to be used to refer – the factions, armies or soldiers involved to them, by twitter users. in the conflict While the first corpus included only tweets that – etc. reference the conflict itself, thus providing a broader view of the online conversation surrounding it, this 3.2.2. Step 5 - Analysis second corpus will include tweets referring specif- From the entities, relations and frames used to char- ically to one or more of the relevant entities and acterize them, which are collected in the previous allow for a more fine-grained analysis. By collecting step, a second, fine-grained and dynamic network these tweets, we aim to identify how the entities will be constructed. This network, which could be 23 Figure 2: Entities network for the Ukraine-Russia conflict. Network filtered by node weighted degree: 0.9999 percentile; and then by edge weight: 0.9 percentile. based of RDF-star or a labelled property graph, is 4. Discussion an in-depth representation of the narratives describ- ing the relations between the key actors and their The network we have already constructed from the characteristics. For this step, we plan to again em- Phase 1 dataset shows an interesting bird’s-eye per- ploy the part-of-speech recognition and dependency spective on the online narratives surrounding the tree identified using the spaCy toolkit to reconstruct selected conflict, that is, the war between Ukraine this network. and Russia. However, this pipeline for automat- ically reconstructing conflict narratives is still at a very early development stage. Open questions remain concerning the advantages and disadvan- tages of the centrality measures and AI methods to be used at each stage of the process, for instance, 24 filtering the actors network. A future aim is to ex- Acknowledgments plore and benchmark these alternative procedures and metrics, as well as their impact on the results, This work was funded by the by the FET-Open throughout future works, employing multiple con- Project #951846 “MUHAI – Meaning and Un- flict datasets. Here follows a brief discussion about derstanding for Human-centric AI” by the EU the criticalities that we have identified at the cur- Pathfinder and Horizon 2020 Program and the rent state of the work and that we will address as a “Empowering Digital Media” grant from the Klaus priority in the next development stages. Tschira Foundation. Firstly, we plan to evaluate the most useful way to construct the co-occurence network. Currently, the network is based on the counts of co-occurences References of identified entities and noun phrases in tweets. [1] A. Whiting, D. Williams, Why people More precisely, the more often two entities appear use social media: a uses and gratifications in the same tweet, the higher the weight of that edge approach, Qualitative Market Research: An is. An alternative approach is to take the frequency International Journal 16 (2013) 362–369. URL: of any two connected nodes into consideration when https://www.emerald.com/insight/content/ having to weigh the relation between two entities doi/10.1108/QMR-06-2013-0041/full/html. (as more frequently occurring entities are also more doi:10.1108/QMR-06-2013-0041. likely to appear together by mere chance). This [2] L. M. Given, D. C. Winkler, K. Hopps-Wallis, metric would also allow us to differentiate between Social Media for Social Good: A Study of pairs of nodes that appear together more often than Experiences and Opportunities in Rural Aus- random, and pairs of nodes that appear together tralia, in: Proceedings of the 8th Interna- less often than random. Similar approaches have tional Conference on Social Media & Soci- been used in other fields of studies, for example, for ety, #SMSociety17, Association for Comput- species probabilistic co-occurrence analysis. In con- ing Machinery, New York, NY, USA, 2017, pp. nection with this, we are also considering a number 1–7. URL: https://doi.org/10.1145/3097286. of different options to calculate the centrality of a 3097293. doi:10.1145/3097286.3097293. node, which is currently based on its degree. [3] T. Mildner, G.-L. Savino, Ethical User In- The second part which we will further investigate terfaces: Exploring the Effects of Dark Pat- is the representation of conflict dynamics over time. terns on Facebook, in: Extended Abstracts The current network is based on the entire set of of the 2021 CHI Conference on Human Fac- tweets collected from a defined time frame. During tors in Computing Systems, Association for the next iteration, we plan to firstly collect tweets Computing Machinery, New York, NY, USA, from a larger time interval and secondly to slice the 2021, pp. 1–7. URL: https://doi.org/10.1145/ network into smaller time-chunks. This will enable 3411763.3451659. us to visualize and analyze how the network changes [4] S. Cunningham, D. Craig, Being ‘re- over time. We expect an interesting perspective ally real’ on YouTube: authenticity, com- to be added to both the co-occurence network of munity and brand culture in social media the first phase as well as to the more fine-grained entertainment, Media International Aus- network. tralia 164 (2017) 71–81. URL: https://doi. Thirdly and finally, we plan for the next iteration org/10.1177/1329878X17709098. doi:10.1177/ of this dataset to analyze more of the context data 1329878X17709098, publisher: SAGE Publica- surrounding the tweets themselves: Many tweets tions Ltd. include metadata about the country of origin, time [5] L. P. Forbes, L. P. Forbes, Does Social Me- zone, location of users, and their language, signify- dia Influence Consumer Buying Behavior? An ing emojis, flags and hashtags, as well as the meta- Investigation Of Recommendations And Pur- data connected to the user’s account. Considering chases, Journal of Business & Economics Re- users’ biographies or geolocation might make it pos- search (JBER) 11 (2013) 107–112. URL: https: sible to look at which narratives are more prominent //www.clutejournals.com. doi:10.19030/jber. among which groups of users or locations. This is v11i2.7623, number: 2. also connected to our previous goal of adding more [6] C. Ellwein, B. Noller, Social Media Min- languages to the corpus. ing: Impact of the Business Model and Pri- vacy Settings, in: Proceedings of the 1st ACM Workshop on Social Media World Sen- 25 sors, SIdEWayS ’15, Association for Comput- How the vulnerability of command-and-control ing Machinery, New York, NY, USA, 2015, pp. systems raises the risks of an inadvertent nu- 3–8. URL: https://doi.org/10.1145/2806655. clear war, International security 43 (2018) 2806656. doi:10.1145/2806655.2806656. 56–99. [7] A. Mathur, A. Narayanan, M. Chetty, Endorse- [16] G. Austin, Unwanted entanglement: the philip- ments on Social Media: An Empirical Study pines’ spratly policy as a case study in conflict of Affiliate Marketing Disclosures on YouTube enhancement?, Security Dialogue 34 (2003) and Pinterest, Proceedings of the ACM on 41–54. Human-Computer Interaction 2 (2018) 1–26. [17] J. Ober, Political conflicts, political debates, URL: http://arxiv.org/abs/1809.00620. doi:10. and political thought, The Shorter Oxford 1145/3274388, arXiv: 1809.00620. History of Europe I: Classical Greece (2000) [8] A. Hermida, F. Fletcher, D. Korell, D. Lo- 111–38. gan, Share, Like, Recommend, Journalism [18] Ö. M. Uluğ, B. Lickel, B. Leidner, Studies 13 (2012) 815–824. URL: https: G. Hirschberger, How do conflict narratives //doi.org/10.1080/1461670X.2012.664430. shape conflict-and peace-related outcomes doi:10.1080/1461670X.2012.664430, among majority group members? the role of publisher: Routledge _eprint: competitive victimhood in intractable conflicts, https://doi.org/10.1080/1461670X.2012.664430. Group Processes & Intergroup Relations 24 [9] K. Holt, A. Shehata, J. Strömbäck, E. Ljung- (2021) 797–814. berg, Age and the effects of news media atten- [19] S. N. Anderlini, V. Gounden, E. Garcia, tion and social media use on political interest B. Harff, M. Khan, D. Khosla, M. S. Lund, and participation: Do social media function D. Nyheim, L. A. Padilla, T. Schmalberger, as leveller?, European Journal of Commu- et al., Journeys through conflict: Narratives nication 28 (2013) 19–34. URL: https://doi. and lessons, Rowman & Littlefield, 2001. org/10.1177/0267323112465369. doi:10.1177/ [20] E. van der Goot, S. Kruikemeier, J. de Ridder, 0267323112465369, publisher: SAGE Publica- R. Vliegenthart, Online and offline battles: tions Ltd. Usage of different political conflict frames, The [10] E. H. R. Rho, M. Mazmanian, Political Hash- International Journal of Press/Politics (2022) tags & the Lost Art of Democratic Dis- 19401612221096633. course, in: Proceedings of the 2020 CHI Con- [21] F. M. Wiant, Exploiting factional discourse: ference on Human Factors in Computing Sys- Wedge issues in contemporary american politi- tems, CHI ’20, Association for Computing cal campaigns, Southern Journal of Communi- Machinery, New York, NY, USA, 2020, pp. cation 67 (2002) 276–289. 1–13. URL: https://doi.org/10.1145/3313831. [22] M. Somer, J. L. McCoy, R. E. Luke, Pernicious 3376542. doi:10.1145/3313831.3376542. polarization, autocratization and opposition [11] J. Aguerri, M. Santisteban, F. Miró-Llinares, strategies, Democratization 28 (2021) 929–948. The fight against disinformation and its [23] A. J. Polsky, Partisan regimes in american consequences: Measuring the impact of politics, Polity 44 (2012) 51–80. “russia state-affiliated media” on twitter, [24] M. Abdel-Fadil, The politics of affect: The OSF Preprints (2022). URL: https://osf.io/ glue of religious and identity conflicts in social preprints/socarxiv/b4qxt/. doi:10.31235/osf. media, Journal of Religion, Media and Digital io/b4qxt. Culture 8 (2019) 11–34. [12] A. Millner, H. Ollivier, Beliefs, politics, and en- [25] G. de Graaf, A. Meijer, Social media and value vironmental policy, Review of Environmental conflicts: An explorative study of the dutch Economics and Policy (2020). police, Public Administration Review 79 (2019) [13] A. Smith, A. C. Stam, Bargaining and the 82–92. nature of war, Journal of Conflict Resolution [26] S. Charap, T. J. Colton, The negative-sum 48 (2004) 783–813. game and how to move past it, Adelphi Series [14] H.-P. Erb, G. Bohner, K. Schmilzle, S. Rank, 56 (2016) 151–184. Beyond conflict and discrepancy: Cognitive [27] M. Dresher, Theory of games of strategy, Tech- bias in minority and majority influence, Person- nical Report, RAND CORP SANTA MONICA ality and Social Psychology Bulletin 24 (1998) CA, 1956. 620–633. [28] C. F. Camerer, Progress in behavioral game [15] J. M. Acton, Escalation through entanglement: theory, Journal of economic perspectives 11 26 (1997) 167–188. doi:10.48550/ARXIV.1904.09131. [29] D. Rohner, M. Thoenig, F. Zilibotti, War signals: A theory of trade, trust, and conflict, Review of Economic Studies 80 (2013) 1114– 1147. [30] A. Hoover Green, The commander’s dilemma: Creating and controlling armed group violence, Journal of Peace Research 53 (2016) 619–632. [31] A. Deighton, The Impossible Peace: Britain, the Division of Germany, and the Origins of the Cold War, Oxford University Press, 1993. [32] U. Can, B. Alatas, A new direction in social network analysis: Online social network anal- ysis problems and applications, Physica A: Statistical Mechanics and its Applications 535 (2019) 122372. [33] T. Willaert, P. Van Eecke, J. Van Soest, K. Beuls, An opinion facilitator for online news media, Frontiers in big Data 4 (2021) 46. [34] K. Beuls, P. Van Eecke, V. S. Cangalovic, A computational construction grammar approach to semantic frame extraction, Linguistics Van- guard 7 (2021). [35] N. Asghar, Automatic extraction of causal relations from natural language texts: a comprehensive survey, arXiv preprint arXiv:1605.07895 (2016). [36] F. van Schalkwyk, J. Dudek, R. Costas, Com- munities of shared interests and cognitive bridges: The case of the anti-vaccination move- ment on twitter, Scientometrics 125 (2020) 1499–1516. [37] C. R. M. A. Santagiustina, M. Warglien, The architecture of partisan debates: The online controversy on the no-deal brexit, PLoS one (2022 (forthcoming)). doi:10.1371/journal. pone.0270236. [38] T. Willaert, S. Banisch, P. Van Eecke, K. Beuls, Tracking causal relations in the news: data, tools, and models for the analysis of argu- mentative statements in online media, Digital Scholarship in the Humanities (2021). [39] S. Gottschalk, E. Demidova, Eventkg–the hub of event knowledge on the web–and biograph- ical timeline generation, Semantic Web 10 (2019) 1039–1070. [40] S. Gottschalk, E. Kacupaj, S. Abdollahi, D. Alves, G. Amaral, E. Koutsiana, T. Ku- culo, D. Major, C. Mello, G. S. Cheema, et al., Oekg: The open event knowledge graph., in: CLEOPATRA@ WWW, 2021, pp. 61–75. [41] A. Delpeuch, Opentapioca: Lightweight entity linking for wikidata, 2019. URL: https://arxiv.org/abs/1904.09131. 27