Modermed: Multimodal Detection of Toxic Behavior in Social Media
Modermed is a research project funded by the Slovak Research and Development Agency that will allow KInIT and the Technical University of Košice to research AI methods for detection of various types of toxic manifestations in multimodal data (text, image, and speech). An important aspect of the project is to ensure that the researched and developed methods will work for Slovak language and that they will be explainable and interpretable.
Antisocial behavior in the online space is in direct contradiction to the intention with which the web and social networks were created – which is to promote the sharing of knowledge and communication between people. To mitigate this behavior, the project focuses on detection of various forms of toxic content on the web, such as:
- offensive, defamatory, or hateful posts,
- posts disrupting or questioning social cohesion,
- disinformation, hoaxes, and manipulative messages.
Modermed aims to address several open problems of existing approaches for toxic content detection, such as lack of approaches capable of processing multimodal data (that is, not only text, but also image or speech or their combinations) and lack of existing available resources (for example labeled datasets) in Slovak for this task. Consequently, the project will:
- research multimodal AI methods for detection of toxic content,
- monitor the development of antisocial and toxic messages and topics,
- recognize the types of authors of toxic content and the inconsistency of their typology with presented content,
- investigate the ways to make the researched models explainable and interpretable,
- ensure the applicability of the models also for the Slovak language and Slovak conditions.
KInIT’s role on the project is to lead the work on applicability of the models for Slovak language, building upon our expertise in the field. We will also contribute to the research of explainability of the detection methods and to the modeling of the topics development. The project is coordinated by the Technical University of Košice; it is thus an ideal opportunity to strengthen the collaboration between KInIT and Slovak academia and to join forces to ensure that future AI will support Slovak language, creating new resources and trained models.
Modermed presents us with an opportunity to collaborate with experts from the Technical University of Košice, building on successful collaborations that we had in the past. Together, we will focus on mitigating the toxic content on the web, which is a problem with a strong impact on society. At the same time, we will also work on strengthening the Slovak NLP resources and models.
Róbert Móro, Senior Researcher
Kempelen Institute of Intelligent Technologies
Project team
Maria Bielikova
Lead and Researcher
Jakub Šimko
Lead and Researcher
Marián Šimko
Lead and Researcher
Miroslav Blšták
Research Engineer
Jaroslav Kopčan
Research Engineer
Róbert Belanec
PhD Student