Social media has become more and more prevalent as our lives become more digitalized, and subsequently online discourse has become integral in important topics like sentiment of public/political figures, consequential health (mis)information spread, and ethics debates. Online discourse can become a helpful tool to spread ideas, facilitate debates and raise awareness, but it can also be a tool of division to further polarize people, spread misinformation, and even influence politics and decisions that governments/establishments make. With this significance, it is important to accurately model online discourse/polarization to better identify relevant actors and groups. Signed networks present a way to better represent and analyze online discourse over traditionally unsigned networks.
Traditional Twitter analysis
The motivation for this paper comes from how limited normal Twitter discourse analysis was limited when using only Twitter Mention and Retweet networks, with a desire to better analyze discourse. The team found that when using Mention networks, members of coalitions would mention both their “allies” in the same coalition along with their “enemies” from other coalitions equally, so there was no distinguishing feature of “friendly” vs “unfriendly” mentions (unless you perhaps count the number of curse words in the tweet). Meanwhile, they found that Retweet networks only signified positive relationships, meaning there were only edges when a user would retweet the tweet of another user. These two networks failed to model an important part of online discourse: the opposition. This served as motivation for the research team: to find a better way to model online discourse (specifically Twitter) to better represent the opposition and negative actors.
The Case: Dutch Tradition and Folklore Character “Black Pete”
The online discourse the research team was studying was Twitter discourse over the tradition of celebrating Dutch folklore character “Black Pete”, a folklore character who has racist characteristics. Specifically, “Black Petes” are the helpers of the Dutch equivalent of Santa Claus, but the controversy comes from the fact that “Black Petes” wear blackface. Using natural language processing and machine learning to analyze and categorize the contents and sentiment of tweets, they represented user interactions as positive and negative edges in the signed network. The first big difference is that they found that 40% of users not included in the retweet network happened to be included the signed network. This is because there were actors who took part in the public discourse of the “Black Pete” topic, but did not happen to retweet or get retweeted, thus being left out of the largest connected component.
An important instance of the retweet networking missing out on significant actors is how it missed the prime minister of the Netherlands Mark Rutte. He did not happen to specifically tweet about the subject, but happened to get mentioned a lot, thus was included in the signed network.
Another big distinguishing feature that signed networks revealed more than just two larger coalitions, and in fact there were many groups expressing a spectrum of support/backlash for/against the tradition. They identified groups that would have positive relationships with other groups while also attacking similar groups on the opposing side, while there were groups that remained isolated and did not attack others.
Source
Keuchenius, A., Törnberg, P., & Uitermark, J. (2021). Why it is important to consider negative ties when studying polarized debates: A signed network analysis of a Dutch cultural controversy on Twitter. PLOS ONE, 16(8). https://doi.org/10.1371/journal.pone.0256696