UMUTEAM at DIPROMATS 2024: Feature Integration for Detecting Fine-grained Propaganda and Narrative
Published in IberLEF 2024, 2024
These notes describe our participation in the 2nd edition of the DIPROMATS shared task, held at IberLEF 2024. This edition repeated the fine-grained detection of propaganda techniques in politics and added an additional subtask for narrative detection, which consists of a multiclass and multi-label classification problem to classify a set of predefined narratives of international actors using few-shot learning. Both tasks are multilingual. For the first task, we propose an approach similar to the one used in the previous edition, combining linguistic features and sentence embeddings using ensemble learning and knowledge integration. For Task 1, we obtain our best result by applying knowledge integration. For the Task 2, we evaluate TuLu and Zephyr, but our results fall below the proposed baseline based on Mixtral 8x7B.
Recommended citation: García-Díaz, J. A., Pan, R., Lázaro, A. R., Cristancho, C., & Valencia-García, R. (2024). UMUTEAM at DIPROMATS 2024: Feature Integration for Detecting Fine-grained Propaganda and Narrative. https://ceur-ws.org/Vol-3756/DIPROMATS2024_paper5.pdf
