TEII: Think, Explain, Interact and Iterate with Large Language Models to Solve Cross-lingual Emotion Detection

TL;DR

The paper tackles cross-lingual emotion detection in Tweets under the EXALT@WASSA2024 task, leveraging multilingual LLMs and multilingual embeddings. It introduces Agentic Workflows that orchestrate multiple LLMs (via Multi-Iteration and Multi-Binary-Classifier variants) and complements them with ensemble voting, yielding strong performance (Ensemble-19 F1=0.6046 on the test set). Key contributions include novel AWF approaches, explanation-focused prompting (ZSE/ZSEC), and an effective LLM selection strategy favoring GPT-4, together with a demonstration that explainability can enhance decision-making in multilingual emotion classification. The results suggest that combining diverse prompting regimes, agentic adjudication, and ensembles can significantly improve cross-linguistic emotion detection, with practical implications for scalable, multilingual sentiment analysis and public opinion monitoring.

Abstract

Cross-lingual emotion detection allows us to analyze global trends, public opinion, and social phenomena at scale. We participated in the Explainability of Cross-lingual Emotion Detection (EXALT) shared task, achieving an F1-score of 0.6046 on the evaluation set for the emotion detection sub-task. Our system outperformed the baseline by more than 0.16 F1-score absolute, and ranked second amongst competing systems. We conducted experiments using fine-tuning, zero-shot learning, and few-shot learning for Large Language Model (LLM)-based models as well as embedding-based BiLSTM and KNN for non-LLM-based techniques. Additionally, we introduced two novel methods: the Multi-Iteration Agentic Workflow and the Multi-Binary-Classifier Agentic Workflow. We found that LLM-based approaches provided good performance on multilingual emotion detection. Furthermore, ensembles combining all our experimented models yielded higher F1-scores than any single approach alone.

Figures (2)

• Figure 1: System architecture of the final ensemble model that combines both individual models and other ensemble models with fewer individual models.
• Figure 2: One iteration of Multi-Iteration Agentic Workflow. (This figure has been designed using images from Flaticon.com)