ARENAS partners at CY Cergy Paris Université team are working on WP 2 in ARENAS, The Characterisation and Detection of Extremist Narratives. Their work specifically addresses Task 2.4 Testing the algorithm with the annotated dataset to validate the annotation schema for automatic detection of extremist narratives; preparing a re-usable model for detecting extremist narratives.
They recently published a paper titled “Analysis of Socially Unacceptable Discourse with Zero-shot Learning”. The paper’s authors are Rayane Ghilene, Dimitra Niaouri, Michele Linardi, and Julien Longhi. Two members of the team Rayane Ghilene and Dimitra Niaouri presented the paper to the 11th International Conference on Computer-Mediated Communication (CMC) and Social Media Corpora. Which was held in Nice, France.
In their presentation, they showed how to tackle the classification of socially unacceptable discourse (SUD) with Zero-Shot Machine Learning. Such a paradigm does not require learning discourse representation from a rigid annotation schema, but it exploits domain knowledge by the possibility of inferring discourse characterisation by drawing hypotheses over textual premises. Such modus operandi provides more flexibility for analysts to leverage a generic context in which models are pre-trained.
It is important to note that SUD is an extensive discourse definition in which Extremist Narrative (EN) can occur. This work aims to define some research directions toward accurately modelling EN Narratological aspects and extremist features.
To read or download the paper presented at the conference click this link.
