Decoding the news: a new application to identify persuasion techniques in the media

Preslav Nakov, professor of natural language processing and chair of the natural language processing department at Mohamed Bin Zayed University of Artificial Intelligence, is working to develop applications that can identify the use of persuasion techniques that are found in the news media. These same techniques that are employed by reporters are also found in disinformation and propaganda campaigns, making Nakov’s work particularly relevant in an age in which it is becoming ever more difficult for people to make sense of the information they are exposed to on the internet.

Nakov’s most recent innovation is called “FRAPPE: Framing, Persuasion and Propaganda Explorer,” an interactive website that allows users to analyze news from around the world. The site provides information about the rhetorical components of news articles and “enables users to gather insights about framing, persuasion and propaganda at an aggregate level, by news outlet and by country,” the researchers write.

The project started about a year and a half ago to provide a tool that helps researchers identify biases in the news. “With FRAPPE you can explore media from different countries, and the application gives you an idea of what the concerns are in those countries about important topics,” Nakov said.

FRAPPE was recently presented at the 18th Conference of the European Chapter of the Association for Computational Linguistics (EACL). The application was developed by an international group of researchers.

The development of the demo started as part of an undergraduate research internship program (UGRIP) in 2023.

The art and the science of persuasion

While Nakov’s research is on the leading edge of innovation in NLP, the study of persuasion has a long history. The ancient Greek philosopher Aristotle has a foundational work on the topic called “Rhetoric” in which he establishes three modes of persuasion. These modes — ethos, pathos and logos — are still used to describe persuasion techniques to this day.

Ethos is based on the credibility, reliability and character of the individual delivering information. Pathos appeals to an audience’s emotions and uses emotional arguments to persuade. Logos relies on reasoning and logic to persuade an audience and includes the use of facts, data and logical arguments to support claims.

Though the three categories of ethos, pathos and logos are used to segment persuasion techniques, there are also more finely grained categories that provide precise descriptions of persuasion methods.

In their study, Nakov and his colleagues use a set of 23 techniques, including appeal to authority, false dilemma or no choice, and conversation killer. The 23 techniques also fall into six broader buckets — attack on reputation, justification, simplification, distraction, calls and manipulative wording.

CAPTION: FRAPPE visualizes 23 persuasion techniques and the way they are used in the media in different countries. Above is a breakdown of the techniques as they are deployed in stories about the Ukraine-Russia war in eight countries. Loaded language is used comparatively less in German news outlets, while name calling is used substantially more when compared to other countries. Flag waving is used more in the USA, UK and Egypt than it is in other countries.

How it’s framed

In addition to analyzing persuasion methods, FRAPPE provides visualization of how news stories are framed by their authors.

Framing is the way media outlets shape the presentation of news stories through their choice of words, images and emphases and it can influence how audiences interpret the information they receive. For example, a story about climate change could be framed as an economic, national security or political issue. A specific story could also include more than one kind of framing technique.

CAPTION: the figure shows how news stories about the Ukraine-Russia war have been framed in different countries. In Belarus, for example, a greater percentage of stories are framed as political than they are in the United States, United Kingdom, United Arab Emirates and Egypt.

“You can see whether certain countries or media emphasize one perspective or another,” Nakov said. “A user can analyze framing to see when a topic like climate change, for example, is framed according to economic, security, resource or political perspective,” he noted.

The researchers trained their system on data from SemEval-2023 task 3, a shared task organized together with the Joint Research Centre of the European Commission, the University of Padova and the Polish Academy of Sciences. The is a dataset consists of approximately 1,600 articles in six languages: English, French, German, Italian, Polish, and Russian. The articles in the data set are annotated according to genre, framings and persuasion techniques. FRAPPE analyzes more than 2 million articles.

In addition to allowing users to compare how news stories are presented in different countries, users can also submit individual stories for analysis, which displays the propaganda and the framing techniques used in that story.

What’s ahead

In the future, the researchers plan to expand the training dataset to include more languages and improve the accuracy of the system to analyze articles in these languages. That said, FRAPPE already processes articles in 100 languages. And while the FRAPPE has analyzed more than 2 million articles, it is currently focused on the two topics of the Ukraine-Russia war and climate change. The researchers will continue to add more articles and more topics that can be examined.

Though persuasive language is not necessarily a bad thing in itself, it can be used to misinform people. And in an age of machine generated content, it may be even more important for people to understand how language can be used to influence their opinions.

“We have seen machine-generated content put online with the purpose to deceive. We see fully automatic websites for fake news. And we see fully automatic fake accounts and some that are coordinated as well,” Nakov said. “And I anticipate more fully automatic multimedia content in the future.”