Hate speech and disinformation have become intractable problems on social media and other online platforms, but there is little agreement on what to do about them. One approach is for companies to monitor and remove hateful or harmful content. Another emerging approach is counter speech, where individual users respond to bullying posts.
But is counter speech an effective strategy to curb online hate and disinformation? It’s a difficult question to address scientifically because so many societal factors are at play beyond the online forums. However, a study published in EPJ Data Science uses a multifaceted approach to begin exploring the question. The study examines four years’ worth of conversations that played out on German Twitter between two self-identified and opposing groups. The results suggest that counter speech may indeed be effective in curbing hateful speech online, especially when done in an organized manner.
Shortly before the 2017 German federal election, a far-right group called Reconquista Germanica began to organize targeted online campaigns, spreading hate and disinformation against immigrants through various social media platforms and promoting a radical-right political party. In April 2018, a counter group called Reconquista Internet organized coordinated counter-messaging.
In their paper, “Impact and dynamics of hate and counter speech online,” Applied Complexity Fellow Joshua Garland and SFI Professor Mirta Galesic, along with former SFI postdoctoral fellow Keyan Ghazi-Zahedi,* Laurent Hébert-Dufresne,** and Jean-Gabriel Young** studied more than 180,000 conversations from 2015 — before the formation of Reconquista Germanica — through 2018.
“This is the first time anyone has done a longitudinal study of complete conversations at this scale,” says Garland. “We were able to collect these conversations and then to study the dynamics between the two groups.”
Because there were self-identified members of both groups, the team was able to train a machine learning classifier to recognize speech patterns typical of hate and counter speech in the conversations.
To get a picture of the effectiveness of counter speech, the authors considered several proxies for effectiveness at multiple scales, from the overall ratios of hate-to-counter speech over time to the dynamics of individual hate and counter-speech posts.
“Across a number of different indicators, we find that organized counter speech appears to contribute to a more balanced public discourse. After the emergence of the organized counter group Reconquista Internet (RI) in the late Spring of 2018, the relative frequency of counter speech increased while that of hate speech decreased,” write the authors.
Similar to what research on “traditional” bullying shows, these findings suggest that the presence of supporting peers can motivate individuals to stand up against online hate speech. “Our work suggests it is important to encourage citizens to stand together against hate and bullying online,” says Galesic. “They will feel empowered, and they can really make a difference.”
The authors are careful to clarify that their study does not identify any causal effects. “There were simultaneous cultural shifts happening in Germany,” says Garland. “We can’t say that organized counter speech caused something to occur, but we can look at the correlation between hate and counter speech.”
While this study focuses on the specific scenario of hate and counter speech on German Twitter, the findings offer insight for addressing other types of online disinformation. “Hateful messaging is really a subset of disinformation,” says Garland. “It’s disinformation about a person or group of people. Our paper shows that organizing matters to fight against disinformation. It might be scary to stand up against a neo-Nazi on my own, but if I can use a hashtag or post on a platform where I have support, it’s easier to stand up against the bully.”
In subsequent work, the authors plan to explore which specific strategies — from humor to counterfactuals to befriending — might be the most effective types of counter speech.
*Keyan Ghazi-Zahedi (Max Planck Institute for Mathematics in the Sciences)
** Laurent Hébert-Dufresne, former James S. McDonnell Fellow at SFI, and and Jean-Gabriel Young (Vermont Complex Systems Center, University of Vermont)
Read the paper, "Impact and Dynamics of hate and counterspeech online," in EPJ Data Science (January 24, 2022). doi:10.1140/epjds/s13688-021-00314-6