It's SUMMER! While I'm attending some virtual conferences and listening to the amazing ideas out there, I'm taking the time to summarize the work that's been out thus far, from my first foray into research. I'm thankful for everyone who's given me a chance, taught me and worked with me.
Case Study Analyses
In the Kashmir Black Day event, where this region is caught in a tussle between India and Pakistan, image analysis from Twitter show bot activity in clusters that call out for action and freedom. The study discovered anomalous actors where many began as food accounts that transformed into promotion accounts.
Indonesia Twitter discourse is very vastly populated, and there are signs of information operation campaigns, which leads to polarization. We formed a pipeline building on the BEND pipeline to characterize information operations by Discovery/ Who/ Did What/ To Whom/ Why/ Impact (under review”).
For Singapore, we developed a method to detect misinformation through skepticism in chat groups (link). Though Twitter is not prevalently used, there are signs of bot activity in recent events, though these bot agents do not influence the conversation (link).
New Techniques
Identification of coordinated clusters with a synchronized action framework where an action (mention, hashtag etc) is performed within a short time interval of each other, evidenced in three templated campaigns.
Where temporal information is not available, coordinating narratives can be found through text similarity magic (“Coordinating narratives in Parler in the Capitol Riots, to appear”) and linear algebra transformation for handling large matrices. Other applications of using text encodings include classifying Twitter misinformation from annotated stories from fact checking websites (link).
Stances on topics like the covid vaccine is variable and agents can flip their stance constantly – but what makes them flip? We construct a social influence model to predict stance flipping behavior through network and past post information (link).
We also model emotion transitions during the early lockdown stages of the COVID19 pandemic using a hidden markov model (link).
Last, there is a characterization of the number of tweets or the number of days required for a stable bot prediction score from available bot detection algorithms (under review).