World

EU-Funded AI Identifies Tens of Thousands of Hate Speech Examples Against Muslims in Finland

According to Mari-Sanna Paukkeri of Utopia Analytics, an AI model to identify hate speech can be built for any language in merely two weeks, if provided a definition of hate speech in a particular culture.
Sputnik

A study commissioned by the Finnish Justice Ministry has discovered nearly 300,000 examples of hate speech posted on Finnish-language websites over a two month period.

The report is based on the analysis of roughly 12 million Finnish-language comments and articles published in September and October 2020. Most of the hate speech discovered (97 percent) was posted on smaller forums and message boards, including one major platform Ylilauta, which is a peer to the commonly known 4chan and is moderated on a volunteer basis.

Trailing message boards, Twitter and Instagram were the next most common platforms for hate speech. Blogs, comments on news websites, and Facebook came in third. Retweets and reposts play a significant role in circulating these messages. For instance, wholly 39 percent of all hate-speech tweets are duplicates.

"Hate speech is not just a challenge for the social media giants. In this report, forums clearly stand out", Miia Aaltonen from the ministry-backed Facts Against Hate project explained.

Of the material discovered, 62 percent was classified as hate rhetoric stigmatising or generalising a group of people, 32 percent as insults, 4 percent as relating to personal characteristics, and 2 percent as directed at a professional group. Remarkably, 26 percent of all comments identified as hate speech featured the word "Muslim", representing a clear tendency.

Furthermore, another significant takeaway is that only a small number of writers appear to produce most of the hate speech. For instance, merely ten individual users accounted for 11 percent of all hate speech identified.

"Everyone is able to influence the spread of hate speech by changing their own behaviour. A small group can get a lot of attention if posts are shared widely", Miia Aaltonen said.

Finnish Top Female Politicians Victimised in Social Media Hate Campaign, NATO Report Claims
According to Mari-Sanna Paukkeri, the CEO of Utopia Analytics that compiled the report, the report relied on artificial intelligence, whose capacity to recognise hate speech depended on the definition provided by human programmers.

"The researchers set out what counted as hate speech and then the material was fed to the artificial intelligence", she explained.

"While the data set consisted of mostly Finnish messages, the results would be very similar in other languages", Paukkeri said. By her own admission, a similar AI model to identify hate speech can be built in any language in merely two weeks. "We only need a skilled individual to say how hate speech should be defined in your culture and language and we need the data to analyse", she said.

The report was commissioned as part of the Facts Against Hatred project funded by the EU. According to the Finnish Justice Ministry, it provides an overall picture of the hate rhetoric that occurs in Finnish public debate on the web.

Discuss