Who does the hateful and insulting speech on social media come from?

To reach this conclusion, the researchers of this study from the University of Sydney, published in Computer Speech & Language, analyzed the content of more than 300,000 tweets from fifteen American political figures. Of the 1,133 tweets classified as hate speech, 1,094 came from right-wing personalities compared to 39 from those on the left. Right-wing personalities contributed to 5,029 of the 5,299 tweets defined as abusive, compared to 270 for those on the left.

The champions of hatred lean to the right…

Among right-wing figures, certain figures stand out for their high number of problematic content. Commenter Ann Coulter, known for her conservative positions, was responsible for nearly half of the hate speech in the dataset, with 464 of the 1,133 tweets classified as such. Former President Donald Trump also made his contribution with 85 messages classified as hate speech and 197 as insulting.

Researchers developed a machine learning model capable of detecting hate speech with increased accuracy, drawing on multiple datasets. The model was trained on eight public datasets, from platforms including Twitter, Reddit and Gab. These datasets varied in their definitions of hate speech, covering aspects such as racism, sexism and general slurs. This way, the tool was better able to detect different types of hate speech.

“There is generally no single definition of hate speech; hate speech is a continuum, as hatred can be expressed overtly using insults and direct references or covertly using sarcasm and even humor“, Marian-Andrei Rizoiu told PsyPost. “Our study develops tools to account for these nuances by leveraging multiple training datasets and a new machine learning technique called transfer learning“.

No monopoly on offensive messages

Among left-wing politicians, Senator Bernie Sanders, former President Barack Obama and former first lady Michelle Obama had no tweets classified as abusive. Representative Alexandria Ocasio-Cortez is the source of four tweets classified as hate speech and four as offensive messages, while her colleague Ilhan Omar had 23 classified as hate speech and 46 as offensive messages.

“What surprised us is that the offensive comments are not only made by right-wing personalities.“, said Marian-Andrei Rizoiu. “Left-wing personalities also spread abusive content in their messages. Although this content would not necessarily be considered hate speech under most definitions, it was abusive.”

Speeches that target the most vulnerable

Hateful content from right-wing figures often targeted specific groups, including Muslims, women, immigrants, and people of color. “We found that most hate tweets targeted topics such as religion (especially Islam), politics, race and ethnicity, women, and refugees and immigrants.,” explains Marian-Andrei Rizoiu.Interestingly, the majority of hate speech was directed towards the most vulnerable groups.”

This study highlights the problem of moderation of social networks and the need to implement more effective tools to moderate hate speech on the platforms. “Detection is only the first step in tackling an online problem“, emphasizes Rizoiu. “The question is how to develop and deploy effective methods in the real online world, which can protect against harmful content without hindering rights such as freedom of expression“.