Unlike many chatbots developing in the space surrounding abuse, our bot is not victim-facing. Our bot is built to interact directly with abusers in order to provoke a conversation from which we will come to better understand the abusers’ motivation in posting abusive content on Twitter.
In order to intervene in these abusive situations safely we looked to previous research that shows us that a profile with more social status, presumably white and male, will likely garner the most engagement from abusive users. Our strategy for interacting with the abuser is to utilize counter speech methods, which are designed to address and condemn the behavior of the abuser, but not the character.
In a preliminary experiment we found that an empathetic tone was the most effective.
The most current version of our bot implements two types of conversations: one neutral and one empathetic.
Our bot will be available on Github and can be used for further research.