Hunting Bots & Trolls in a Game Environment

“Hunting Bots & Trolls in a Game Environment“

Speaker: Shulman, S. (Texifter)

Date: 12 February 2020

Speaker Session Preview

SMA hosted a speaker session presented by Dr. Stuart Shulman (Texifter) as a part of its SMA General Speaker Series. Dr. Shulman’s brief focused on the ways in which one identify Internet bots and trolls by using text classification programs. He stated that though machines enable one to classify large bodies of text, it remains difficult for machines to be consistently accurate; therefore, a human must remain in the loop. Dr. Shulman then spoke about CoderRank, a program that ranks humans based on trust and knowledge vectors. The idea behind the program is, when training machines for text analysis, greater reliance should be placed on the specific inputs of those humans most likely to create a valid observation. CoderRank, therefore, is designed to identify those who are most qualified to code and classify text. Dr. Shulman proceeded to discuss lessons learned from tracking down fake accounts in the Trudeau and Biden campaigns. In the Trudeau campaign, Dr. Shulman and his team looked for biographical indicators that were not Canadian, conducted searches based on keywords, and analyzed abnormal text. In the Biden campaign, Dr. Shulman and his team looked for accounts with significant followings that had, in reality, purchased social media followers. He stated that as we move into this next technological landscape, it is only going to become more difficult to discern who is real and who is not online. Moreover, if one immerses oneself in social media profile data and looks for specific indicators, one can see a mixture of real, authentic people; human-assisted bots; bot-assisted humans; and pure bots. Dr. Shulman then spoke about bot detection based on the frequency of tweets. He and his team of coders create filters, labels, and other features in order to determine where the tweets are coming from. They also examine Twitter accounts’ networks to decipher whether or not the accounts of interest are operated by humans. Dr. Shulman expressed, however, that gaining access to data is becoming increasingly difficult due to increased regulations and data privacy restrictions, which makes it harder for him and his team to identify bots. To conclude, Dr. Shulman spoke about how his text classification and bot detecting techniques can be applied to a gaming environment, specifically with regards to veteran suicide prevention and school shooters prevention.

‍