Social media sites such as Twitter have faced significant pressure to mitigate spam and abuse on their platform in the aftermath of congressional investigations into Russian interference in the 2016 U.S. presidential election. Twitter publicly acknowledged the exploitation of their platform and has since conducted aggressive cleanups to suspend the involved accounts. To shed light on Twitter's countermeasures, we conduct a postmortem analysis of about one million Twitter accounts who engaged in the 2016 U.S. presidential election but were later suspended by Twitter. To systematically analyze coordinated activities of these suspended accounts, we group them into communities based on their retweet/mention network and analyze different characteristics such as popular tweeters, domains, and hashtags. The results show that suspended and regular communities exhibit significant differences in terms of popular tweeter and hashtags. Our qualitative analysis also shows that suspended communities are heterogeneous in terms of their characteristics. We further find that accounts suspended by Twitter's new countermeasures are tightly connected to the original suspended communities.

There is a growing concern about the extent to which algorithmic personalization limits people's exposure to diverse viewpoints, thereby creating “filter bubbles” or “echo chambers.” Prior research on web search personalization has mainly reported location-based personalization of search results. In this paper, we investigate whether web search results are personalized based on a user's browsing history, which can be inferred by search engines via third-party tracking. Specifically, we develop a “sock puppet” auditing system in which a pair of fresh browser profiles, first, visits web pages that reflect divergent political discourses and, second, executes identical politically oriented Google News searches. Comparing the search results returned by Google News for distinctly trained browser profiles, we observe statistically significant personalization that tends to reinforce the presumed partisanship.

The American Voter - a seminal work in political science - uncovered the multifaceted nature of voting behavior which has been corroborated in electoral research for decades since. In this paper, we leverage The American Voter as an analysis framework in the realm of computational political science, employing the factors of party, personality, and policy to structure the analysis of public discourse on online social media during the 2016 U.S. presidential primaries. Our analysis of 50 million tweets reveals the continuing importance of these three factors; our understanding is also enriched by the application of sentiment analysis techniques. The overwhelmingly negative sentiment of conversations surrounding 10 major presidential candidates reveals more "crosstalk" from Democratic leaning users towards Republican candidates, and less vice-versa. We uncover the lack of moderation as the most discussed personality dimension during this campaign season, as the political field becomes more extreme - Clinton and Rubio are perceived as moderate, while Trump, Sanders, and Cruz are not. While the most discussed issues are foreign policy and immigration, Republicans tweet more about abortion than Democrats who tweet more about gay rights than Republicans. Finally, we illustrate the importance of multifaceted political discourse analysis by applying regression to quantify the impact of party, personality, and policy on national polls.