Until recently, social media were seen to promote democratic discourse on social and political issues. However, this powerful communication ecosystem has come under scrutiny for allowing hostile actors to exploit online discussions in an attempt to manipulate public opinion. A case in point is the ongoing U.S. Congress investigation of Russian interference in the 2016 U.S. election campaign, with Russia accused of, among other things, using trolls (malicious accounts created for the purpose of manipulation) and bots (automated accounts) to spread propaganda and politically biased information. In this study, we explore the effects of this manipulation campaign, taking a closer look at users who re-shared the posts produced on Twitter by the Russian troll accounts publicly disclosed by U.S. Congress investigation. We collected a dataset of 13 million election-related posts shared on Twitter in the year of 2016 by over a million distinct users. This dataset includes accounts associated with the identified Russian trolls as well as users sharing posts in the same time period on a variety of topics around the 2016 elections. We use label propagation to infer the users' ideology based on the news sources they share. We are able to classify a large number of the users as liberal or con-

In recent years, social media has revolutionized how people communicate and share information. One function of social media, besides connecting with friends, is sharing opinions with others. Micro blogging sites, like Twitter, have often provided an online forum for social activism. When users debate about controversial topics on social media, they typically share different types of evidence to support their claims. Classifying these types of evidence can provide an estimate for how adequately the arguments have been supported. We first introduce a manually built gold standard dataset of 3000 tweets related to the recent FBI and Apple encryption debate. We develop a framework for automatically classifying six evidence types typically used on Twitter to discuss the debate. Our findings show that a Support Vector Machine (SVM) classifier trained with n-gram and additional features is capable of capturing the different forms of representing evidence on Twitter, and exhibits significant improvements over the unigram baseline, achieving a F 1 macroaveraged of 82.8%.

The ease with which information can be shared on social media has opened it up to abuse and manipulation. One example of a manipulation campaign that has garnered much attention recently was the alleged Russian interference in the 2016 U.S. elections, with Russia accused of, among other things, using trolls and malicious accounts to spread misinformation and politically biased information. To take an in-depth look at this manipulation campaign, we collected a dataset of 13 million election-related posts shared on Twitter in 2016 by over a million distinct users. This dataset includes accounts associated with the identified Russian trolls as well as users sharing posts in the same time period on a variety of topics around the 2016 elections. To study how these trolls attempted to manipulate public opinion, we identified 49 theoretically grounded linguistic markers of deception and measured their use by troll and non-troll accounts. We show that deceptive language cues can help to accurately identify trolls, with average F1 score of 82% and recall 88%.

Social media have enabled a revolution in user-generated content. They allow users to connect, build community, produce and share content, and publish opinions. To better understand online users' attitudes and opinions, we use stance classification. Stance classification is a relatively new and challenging approach to deepen opinion mining by classifying a user's stance in a debate. Our stance classification use case is tweets that were related to the spring 2016 debate over the FBI's request that Apple decrypt a user's iPhone. In this "encryption debate," public opinion was polarized between advocates for individual privacy and advocates for national security. We propose a machine learning approach to classify stance in the debate, and a topic classification that uses lexical, syntactic, Twitter-specific, and argumentative features as a predictor for classifications. Models trained on these feature sets showed significant increases in accuracy relative to the unigram baseline.

No abstract