Lotus at SemEval-2025 Task 11: RoBERTa with Llama-3 Generated Explanations for Multi-Label Emotion Classification

TL;DR

This work introduces Lotus, a two-phase framework that fuses LLaMA-3-generated explanations with RoBERTa to tackle multi-label emotion classification on the BRIGHTER dataset. By enriching text with contextually rich explanations, the approach achieves a Macro F1 of 0.7396 (±0.0016) and Micro F1 of 0.7678, outperforming a text-only RoBERTa baseline and demonstrating notable gains for emotions such as Fear, Joy, and Sadness. While not top-ranked on the SemEval task, the results validate that explanatory content helps resolve ambiguity and improve multi-label detection across nuanced emotional cues, with future work targeting underrepresented emotions and multilingual generalization. The study contributes a concrete methodology for integrating generated explanations into transformer-based emotion detectors and provides reproducible code resources for further exploration.

Abstract

This paper presents a novel approach for multi-label emotion detection, where Llama-3 is used to generate explanatory content that clarifies ambiguous emotional expressions, thereby enhancing RoBERTa's emotion classification performance. By incorporating explanatory context, our method improves F1-scores, particularly for emotions like fear, joy, and sadness, and outperforms text-only models. The addition of explanatory content helps resolve ambiguity, addresses challenges like overlapping emotional cues, and enhances multi-label classification, marking a significant advancement in emotion detection tasks.