Machine Learning Model

Our research team developed a machine learning model to classify accounts engaged in toxic trolling or harmful inauthentic activity. Due to the subjectivity of identifying such behavior, we adopted a unique approach. Instead of relying on our subjective interpretation, we used Twitter rules as a guide to identify accounts repeatedly violating the platform's rules.

We gathered data by searching for accounts that engaged in behaviors such as hate speech, harassment, impersonation, and spamming, among others, which violated the rules. We then trained our model on this dataset to classify accounts exhibiting similar patterns of behavior as "problematic."

Our model's classification process is based solely on behavioral patterns, not on factors such as ideology, political affiliation, religious beliefs, geographic location, or frequency of tweets. Therefore, our model's classification is unbiased and purely data-driven.