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

2025-02-27

AI Summary

Fine-grained emotions—such as fear, joy, and sadness—in multi-label sentiment classification suffer from high ambiguity, semantic overlap, and label correlation, leading to poor discriminability and interpretability. Method: We propose an explainability-enhanced framework that explicitly integrates semantically rich, Llama-3-generated explanatory text into the classification pipeline. This explanation clarifies ambiguous emotional expressions by providing contextual grounding; it is jointly encoded with the original input via a fine-tuned RoBERTa encoder and optimized end-to-end using multi-label classification loss. Contribution/Results: Evaluated on SemEval-2025 Task 11, our method achieves significant gains in macro-F1, particularly for highly ambiguous classes (e.g., fear, joy, sadness), demonstrating both improved predictive accuracy and enhanced model transparency. By bridging generative explanation with discriminative classification, the framework establishes a novel paradigm for interpretable multi-label sentiment analysis.

Technology Category

• Natural Language Processing: Text Classification & Sentiment Analysis
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Interpretability, Explainability, and Transparency

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

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.

Problem

• Enhances RoBERTa's emotion classification
• Resolves ambiguity in emotional expressions
• Improves multi-label classification performance

Innovation

• RoBERTa with Llama-3 explanations
• Multi-label emotion classification improvement
• Enhanced F1-scores for emotions