How one NGO is using AI to change how antisemitism is monitored

A Hungarian NGO is using an artificial intelligence-based web scraping mechanism to monitor antisemitism and sentiments on social media, with the objective of creating a monitoring system around Europe with unified metrics.

Created in 2012 in a collaboration between Hungarian Jewish communities and civil society groups, the Action and Protection Foundation (TEV) began operating in 2013, with one of its tasks being the monitoring of antisemitic activity in the country.

Many Jewish organizations around Europe reported antisemitic activity through community reports, but TEV Chairman Kalman Szalal felt that monitoring helped better understand antisemitism and thereby guide objectives, such as improving Holocaust education or pursuing new legislation.

For almost a decade, TEV monitored radio, television, and newspapers, but three years ago it realized this was not enough, as mainstream media was already largely self-regulated, and much of the world’s zeitgeist could only be tapped into on social media.

Szalal, previously working in software development, worked with his team to develop a tool to fit their purposes. They trained AI on antisemitic materials to identify the best keywords and then developed AI-based scraping software to scan Facebook, Instagram, YouTube, LinkedIn, and Reddit.

The system gathers social media posts, prioritizing them by likes, comments, shares, and the likelihood that, in the context, they are truly antisemitic. A brief summary is given of the text’s topics.

At this point, a human analyst is still needed to verify that the text is indeed antisemitic and to categorize the type of content as Holocaust denial, conspiracy theories, anti-Judaism, structural antisemitism, classical antisemitism, or anti-Israel.

According to Szalal, in the near future, human analysts will no longer be needed, as automation will allow for deeper analysis of antisemitic trends.

Based on the data, Szalal’s analyses show that in Hungary, online anti-Israel sentiment has become slightly more popular than classical antisemitism – though the latter is more common among male social media users, while the former is more widespread among women.

Women were often more swayed by the emotional images coming out of Gaza, according to TEV.

Szalal believes that anti-Israel rhetoric often disguises antisemitism and comes from a slow osmosis of Western European ideology and sentiment into Hungary.

His research also showed that while support for Israel’s right to defend itself was growing, there was also a trending opinion that Israel was committing a genocide in Gaza.

The system doesn’t treat all anti-Israel opinions as antisemitic, Szalal insisted, but has built in the International Holocaust Remembrance Alliance working definition of antisemitism to determine when criticism crosses into the territory of demonization.

Younger demographic groups also preferred anti-Israel manifestations of antisemitism online, according to TEV’s statistics.

Szalal assessed it was natural that younger groups who grew up online would use different parlance and would want to distance themselves from more classic and controversial terms and phrases associated with Jew-hatred.

The software has no language limit and is currently operating in France, Germany, and Sweden. Szalal wants to proliferate its use in other European countries and to create common systems and definitions for measuring antisemitism.

As it stood, with different systems for reporting and monitoring in each state, it was difficult to compare the sentiments in different countries, and thereby harder to ascertain the correct course of action, such as legislation.

The data collected by TEV helped the organization provide legislative support and, in some cases, refer incidents such as incitement to prosecutors or the police. Szalal explained that legislation – like a 2010 law – has been an effective tool in deterring Holocaust denial.

TEV’s September data showed that only 0.01% of the posts it collected in Hungary contained Holocaust denial.

Understanding the rhetoric used in antisemitism also informed education, such as a high school program in which a rabbi presents his Jewish identity and opens the floor to any student questions.

Engaging with the intellectual and emotional sensitivities of students would ostensibly help them understand facts and historic events. TEV has also been involved in revising materials in 32 elementary school texts and 45 secondary school texts, with 70 and 50 of the proposed amendments adopted, respectively.

The benefits of understanding, developing, and comparing tools and methodology for other countries are clear, though Szalal acknowledged the system would have to be fine-tuned to each polity’s legal and privacy restrictions.

Szalal has been campaigning to see the system adopted by other Jewish organizations and communities, but there have been hurdles. There are high costs involved in adopting monitoring systems, setting up the required equipment, and hiring human resources.

It is essential that local communities work on the project so they don’t feel that outside actors are determining what antisemitism is.

Szalal suggested that NGOs or even the State of Israel could be involved in finding solutions for the costs and challenges faced by local communities in adopting the system.

TEV said in a 2024 analysis of survey data from the European Union Agency for Fundamental Rights (FRA) that Hungary was one of the worst performers in terms of antisemitic harassment in 2013 but that it became one of the top performers by 2018.

While 90% of Hungarian Jews saw antisemitism as a major problem in 2013, the concern had subsided by 13% within five years. The FRA said that 65% of respondents saw antisemitism as a problem in Hungary in 2023.

Szalal said there is an odd reality that dictates online and offline antisemitism.

In Eastern Europe, there is often a lot of prejudice expressed, but there are few antisemitic incidents; the reverse is true for Western Europe.

He assessed there might be a distortion of the information, in which people may not have vocally expressed their true opinions in the West because it is not politically correct.

There are also far more Middle Eastern migrants in the West, and Israel is more involved in incidents there than in the East.

TEV offered its data-centric approach as a means to better understand such a phenomenon, arguing that one could only hope to fight antisemitism if one first understood antisemitism.