The DEMON (Detect and Evaluate Manipulation of ONline information) project aims to characterize and detect manipulations of online information by combining both Natural Language Processing (NLP) techniques and social network analysis.
From the NLP perspective, different variations in language, such as certain utterances, can be exploited to understand the different ways in which the same news may be expressed (and perceived) by readers across online channels. From a second perspective, network-based techniques can be used to assess the role and influence of sources and actors in manipulating specific information for different purposes.
This project will focus on using advanced language models (LLMs) to refine the analysis of online information, enabling a deeper understanding of the subtexts and linguistic nuances used in information manipulation. The use of LLMs will add an additional dimension to our ability to detect and understand the subtle tactics employed by malicious actors in disseminating manipulated information through online channels.
The novelty of the proposed approach lies in the fact that it departs from most research that seeks to automatically detect false information by
- introducing and adopting a categorization of the sub-narratives that characterize different types of online information
- leveraging it to detect the spread of weaponized and manipulated information in online social media.
The outcome of this project will advance the state of the art in understanding various aspects of potentially harmful online information by combining approaches based on recent NLP analysis applied to content with methods using data science and network science.
The project will also have a substantial impact in detecting and reducing the effects of potentially dangerous cyber attacks on public discussions at the national level conducted by foreign entities and coordinated groups of malicious actors, and contribute to the current debate on content moderation and intervention on online social platforms.
