Mapping Ideological and Emotional Landscapes in Online Far-Left Discourse: A Computational Analysis

This study examines the ideological and emotional mechanisms underpinning far-left discourse in Greece during a transformative period of political and economic upheaval. Utilizing a corpus of 308,327 posts from a prominent online hub for grassroots activism and radical leftist ideologies, we aim to construct a comprehensive narrative of ideological positioning and emotional expressions within this influential digital sphere. Our approach combines advanced closed-source models (GPT-4o-mini) with open-source Greek language models (Meltemi) as well as state-of-the-art open-source models (DeepSeek-v3) to annotate the dataset. These annotations will inform the development of a supervised classification model to systematically analyze the corpus. By focusing on both basic emotions (e.g., anger, fear, hope) and complex mechanisms (e.g., victimhood, political efficacy, nostalgia), we seek to identify how ideological and emotional narratives evolved in response to events such as the Greek economic crisis. Our pilot analysis demonstrates the efficacy and cost-effectiveness of large language models in a low-resource linguistic setting, while also highlighting challenges related to consistency and intercoder reliability. Additionally, we present performance metrics benchmarked against human annotators to assess the validity of our approach. Our research provides novel insights into the interplay between online discourse and real-world political behavior in digital spaces.