Automatic Recognition of human-written and deepfake-generated text disinformation in social media for a low-resourced language

Aim

The aim of the project is the automatic detection of TRACES of intentionally spread textual disinformation, written in Bulgarian in social media. The project tackles both intentional disinformation written by humans and such, generated via deepfake methods.

Research

Interdisciplinary methods will be applied. The methods will be drawn from the following fields: Natural Language Processing, Linguistics, Psycholinguistics, Machine Learning, Journalism.

The research will include:

1. Determining and systematizing a large number of linguistic and psycholinguistic markers of human lies valid for the Bulgarian language.
2. Collecting new large-scale Bulgarian-language datasets from social media on the topics of “Covid-19” and “politics”.
3. Manual annotation by journalists and fact-checkers.
4. Implementing and discovering the most appropriate machine learning methods for detecting human intentional and deepfake disinformation in social media in the Bulgarian language.

More information about the project is available at https://traces.gate-ai.eu/