Seminar at IPI PAN: "The Why and How of Disinformation: Datasets, Methods, and Language Models Evaluation"

Arkadiusz Modzelewski is also a PhD student in a cotutelle program at the University of Padua and the ICT & Design Doctoral School run by the Polish-Japanese Academy of Information Technology

In his talk, titled “The Why and How of Disinformation: Datasets, Methods, and Language Models Evaluation”, he presented the results of his PhD research and discussed his recent publications on disinformation, persuasion, and the malicious intent behind misleading information.

During the seminar, Arkadiusz introduced MIPD, one of the largest Polish disinformation datasets, created to analyze disinformation in terms of malicious intent and manipulative techniques. He also presented a novel approach to improving disinformation detection through persuasion knowledge, using the Persuasion-Augmented Chain of Thought method . In addition, he discussed conducted research on disinformation narratives, specifically he introduced the EU DisnfoTest benchmark for evaluating large language models and DiNaM, a method for extracting disinformation narratives, which also will be soon published at EMNLP 2025.

The seminar provided participants with in-depth knowledge about the complex relationships between disinformation, information manipulation, and the intentions of entities responsible for spreading false content. During the event, mechanisms behind the creation of disinformation were discussed, along with current approaches and directions aimed at improving the ability of large language models to identify such content.

We are pleased that our employee had the opportunity to present these issues during the seminar organized by the Polish Academy of Sciences, which is a source of pride for us and a confirmation of the value of our research and development efforts.