A Swedish company is using AI to measure emotions in news about Swedish politics

It is election year in Sweden, and the media is filled with news articles about the eight major parties as well as individual politicians. Leading up to the election, Mynewsdesk, a Swedish company specialising in communications technology, has used AI to find out which emotions are hidden in the news reporting, and the results show that a majority of news about Swedish politics are not just negative but also emotionally charged.

Mynewsdesk has analysed over 20 000 articles from more than 600 digital sources, and the analysis shows that over 70 percent of the news are categorised as “negative” by the AI. The result also allows you to see which parties receive the most negative vs. positive news and which leaders are most intensively reported on. The possibilities offered by AI technology create new conditions for monitoring the media environment and for media analysis, which may benefit the general public in the future. Not least when it comes to major issues such as democracy, source criticism, and the political debate climate, according to Louise Barnekow, CEO at Mynewsdesk.

– What I find most interesting is not the individual differences between the parties in the reporting but the bigger picture. In our analysis, we clearly see that many articles about the Swedish election are negative and emotionally charged. What effect does that have on those following and reading the news? And how does this impact the polarisation in our society? These are the types of questions that Mynewsdesk’s analysis highlights. Another thing that I find interesting is what role AI technology, such as the one we apply here, can play for journalism in the future and for the democratic conversation as a whole, says Louise Barnekow.

To carry out the analysis, Mynewsdesk has used a technique called 'transfer learning’. By using a comprehensive language model and letting it assess a text's intensity and emotional state, the algorithm has swept the internet for all news related to Swedish politics and analyses individual words and abbreviations, synonyms, sentences, and context. This makes the method very sophisticated, says Ola Gustafsson, Data Scientist at Mynewsdesk.

– Ahead of the 2022 election, Mynewsdesk has used modern language technology to define the emotional state of news reporting. We can see whether the text is positive or negative, but also the intensity with which the feeling is conveyed, high or low, upset or calm. The ability to identify the intensity also provides clues for which texts and topics are polarising. We have gone through all of the online news articles about Swedish politics since the first of June and intend to analyse news about Swedish politics and the autumn elections on an ongoing basis. The results are incredibly interesting, says Ola Gustafsson.

In the future, AI like the one Mynewsdesk uses to analyse news could be used to help news consumers with source criticism, for example. When it comes to source criticism, the focus is often on what is said and by who, not how it is said. By utilising an AI algorithm that can measure the intensity and analyse news texts, it would be possible to obtain a type of ‘product declaration’ for news, which could help readers judge whether a text is polarising or loaded with political values.

That the media, and society in general, is becoming increasingly polarised is something that has been discussed for a long time, and which becomes even more relevant during an election year. With the help of AI, it is possible to show in black and white what it actually looks like.