New tool maps and analyzes online hate speech

An advanced tool has been developed to map and analyze online hate speech, addressing a growing concern over the rapid increase in such behavior on social media. Recently, the number of reported hate crimes in Norway has more than quintupled since 2012, with around 40% of these incidents being hate speech, primarily targeting skin color, ethnicity, or sexual orientation.

Notably, politicians and public commentators are also increasingly subjected to threats and harassment. This development is attributed to the rapid evolution of digital platforms, where hate speech can spread more quickly and reach a broader audience.

To combat this issue, researchers at NTNU and collaborators from various institutions have created an innovative tool. This tool is not solely designed for tracking and prosecuting individuals but also aims to identify environments that require targeted interventions, thus helping the police and municipalities allocate resources more effectively.

The tool features a function that connects individual instances of hate speech, creating a visualized network of statements to illustrate their origins, dissemination, and how discussions evolve over time.

The researchers emphasize the complexity of hate speech, as its definition varies legally and culturally from country to country and person to person. To overcome this challenge, they identified four central characteristics: ethnicity, gender identity, religion, and sexual orientation. These categories are not only prevalent in Norway but also align with the U.N. International Covenant on Civil and Political Rights. Consequently, the tool proves applicable beyond Norway's borders.

When the tool identifies content that falls into one or more of these categories, it assesses the content's toxicity level, ranging from neutral to potentially offensive, hateful, or aggressive. This assessment relies on machine learning algorithms trained on a substantial dataset of 14,000 sentences from platforms like Facebook, Reddit, and Twitter. The researchers ensured transparency by explaining the factors contributing to the tool's categorization of statements as hateful, fostering trust in the technology.

Moving forward, the researchers aim to enable the tool to detect hate speech in real-time by automatically retrieving data from digital platforms instead of relying on manual input. As the tool becomes more automated and integrated into continuous development, it will improve its accuracy and effectiveness through human feedback and ongoing learning. However, the researchers remain committed to maintaining human oversight to ensure the tool's reliability and ethical standards.