Analysing more than a decade of texts

The team – consisting of João Fonseca, Gossa Lô, Erman Acar and Fernando Santos – analysed more than a decade of Dutch-language texts from three domains: transcripts from the Tweede Kamer, articles from national newspapers, and comments posted on major Dutch news channels on YouTube. They focused on seven legally recognised grounds of discrimination. In total, the dataset covers millions of comments and texts, each linked to a specific point in time. Their open-source pipeline combines keyword analysis, sentiment analysis, large language model (LLM) classification and time-series methods (including Granger causality) to detect trends and cross-domain influence.

Online comment streams reveal rising activity around antisemitism, Islamophobia and origin

The results show that discriminatory discourse is not confined to the fringes of the internet. Some of the most sensitive forms of discrimination – antisemitism, Islamophobia and discrimination based on origin – appear frequently and increasingly in online comment sections, especially around news from certain outlets. For example, there has been a rising volume of posts related to origin and Islamophobia on De Telegraaf’s YouTube channel since 2019, and in the NOS comment section in 2024.

Discriminatory language increasingly present in mainstream public debate

Crucially, the study finds that offline discourse can shape what happens online – and vice versa. Parliamentary debates in the Tweede Kamer frequently precede shifts in discriminatory discourse on YouTube and in newspapers, while online discussions sometimes anticipate later parliamentary attention to sensitive topics. The mention of certain political parties in the news is statistically linked to subsequent increases in potentially antisemitic, Islamophobic and origin-related expressions in YouTube comments, suggesting that discriminatory narratives are increasingly embedded in mainstream public arenas.

From evidence to action: guidelines and data-driven tools

Building on these findings, the authors argue that discrimination should be approached as a dynamic “climate” rather than a series of isolated incidents. They call for integrated, near real-time monitoring systems that combine political, media and online data to provide early warning signals of escalating discriminatory discourse, especially on sensitive grounds such as antisemitism, Islamophobia and origin.

The report also recommends explicit norms and behavioural guidelines for the Tweede Kamer to safeguard the “soft guardrails” of democracy, stressing that what is said in parliament can ripple outward into broader society. Finally, the researchers urge policymakers, watchdogs and civil society organisations to use data-driven tools to design targeted interventions and public awareness campaigns that counter harmful dynamics before they become entrenched.

Enquiries and requests

For press related enquiries please contact Femke van Zijst of the UvA Press Office