.png)
Two of the VIGILANT project consortium members, the Kempelen Institute of Intelligent Technologies (KInIT) and the University of Sheffield proposed the winning solutions at the SemEval 2023 data challenge – the most prestigious data challenge within the domain of Natural Language Processing (NLP).
The challenge comprised 12 diverse tasks spanning various application domains. The focus of the winning solutions was on predicting the credibility of text within web articles, aligning closely with the Vigilant project's objective of aiding law enforcement in identifying malicious online content.
As part of the data competition, KInIT concentrated on the persuasion technique detection subtask, while the University of Sheffield tackled two subtasks: news genre categorisation and framing detection. Using a proposed AI solution, the KInIT team secured the top position in six out of nine languages, including Italian, Russian, German, Polish, Greek, and Georgian, with notable success in two previously unseen languages. Simultaneously, the University of Sheffield claimed the second spot for English.
The key to their success lays in navigating the challenge of multilinguality. Initially exploring a monolingual model operating solely in English, they later adopted a fully multilingual model capable of understanding and working with all languages. Utilizing pretrained large language models, such as RoBERTa and XLM-RoBERTa, fine-tuned specifically for persuasion technique detection, the multilingual solution proved superior, ultimately becoming the chosen method for the final solution.
Encouraged by their accomplishments, KInIT remains committed to further experimentation, including probing state-of-the-art large language models such as ChatGPT and GPT-4, with expectations of yielding new insights and deploying high-performing models in credibility assessment services.
This achievement underscores the collective success of both teams in the complex task of credibility detection, contributing significantly to the VIGILANT project's mission. The positive results serve as a new stimulus for ongoing research with the potential of integrating these winning models into the solutions and tools developed within the project.
Learn more about the winning solutions in the papers summarizing the approach taken by KInIT and the University of Sheffield for their respective subtasks.