A key part of politics is creating an individual voice, brand, and affiliation. In the last decade, we have started to see a shift in the methods used to create a politician’s brand. Social media platforms, particularly X (formerly known as Twitter), have to be carefully curated, as they become a reflection of the politician’s stances, supporters, and proponents.
Most political strategies take social media into account, and political moves are also hosted on social media platforms. With these advancements in mind, Benjamin Leinwand, Assistant Professor of Mathematical Science at Stevens, was able to use social media politics as an application of a new model.
The researchers analyzed 475 members of Congress. The members analyzed had at least 10 tweets between the dates of February 9 and June 9, 2022. The model was not given any information on the members’ political affiliations, status, or party. All members were organized into three groups by the model, all of which fell along familiar political affiliations. The three groups can be defined as the following categories: (1) Senators, (2) Democratic Congresspeople, and (3) Republican Congresspeople. Person A and Person B were defined as connected if one tweeted at the other or retweeted the other person.
The researchers needed a method to map out the vast and complex web of online connections. Typically, researchers use methods developed for sparse and unweighted networks. A sparse network, in the context of social media, is one in which most users are not connected — users don’t often follow or interact with each other. An unweighted network only takes into account whether a connection exists — a binary valuation, rather than a strength or frequency of a connection.
This research method varied from the norm the most in the fact that the model used was conceptually based on a dense and weighted network, which had to then be reworked in order to fit the Twitter dataset. This allowed for more flexibility of the types of data the model can handle as well as its outcomes. The model does not assume that every aspect of the network has the same mathematical value or pattern, therefore allowing the structure and data itself to shape the resulting predictions. Though it may seem trivial, “differences in the underlying process as well as what gets measured can change where the information lives in a network,” as explained by Professor Leinwand, which ultimately impacts the findings. The foundation of the method used also ensured that all resulting values were probabilities between zero and one.
The members most often tweeted with people within their group (Democratic congresspeople interacted with other Democratic Congresspeople, Senators interacted with other Senators, and similarly with Republican Congresspeople). Democratic Congresspeople were more likely to interact with other groups as compared to their Republican counterparts. An explanation for this could be that Democrats had control of the House at the time, so they may have been more incentivized to branch out.
Interacting within groups was the general outcome that 463 out of the 475 followed. However, there were 12 nonconforming individuals. Based on their patterns of interaction, these 12 individuals acted as if they were a part of another group. They were not simply interacting with those within their group, but with Senators as well. Due to their interactions, they were grouped inaccurately, classified as parts of groups they were not a part of. Of the 12 nonconformists, two ended up being elected senators, as in two of the 12 later became a part of the group the model classified them as. A primitive conclusion of this fact is that individuals may appear more senatorial depending on the amount of interactions, though Leinwand says, “if I had to guess, social media positioning is more a symptom than a cause,” and to arrive at a formal conclusion, more research would be needed.
The researchers establish that this is a beginning to the research, and to derive more conclusions about political strategy and science, more information will be needed. When asked about next steps in this subject, Professor Leinwand proposed looking at the data through a sentiment analysis, looking at the content of the tweets more than just the interactions. In terms of the model itself, Leinwand hopes to “create and model even more flexible and expressive networks.”
