Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

JNLP at SemEval-2025 Task 11: Cross-Lingual Multi-Label Emotion Detection Using Generative Models

Jieying Xue, Phuong Minh Nguyen, Minh Le Nguyen, Xin Liu

TL;DR

This work tackles cross-lingual, multilingual multi-label emotion detection for SemEval-2025 Task 11 by integrating a fine-tuned BERT-based classifier with instruction-tuned generative LLMs. It investigates two multi-label strategies, base (predict all labels) and pairwise (one-label-at-a-time), and examines both Track A (emotion detection) and Track B (emotion intensity) across multiple languages, including low-resource ones. Mixed-language training and the choice between BERT-based and LLM-based architectures are analyzed, with findings showing when each approach excels and how the strategies interact with label distribution and data sparsity. The approach achieves strong multilingual generalization, securing top placements in several languages and providing open-source code for reproducibility, which has practical implications for real-world cross-lingual emotion analysis in social media.

Abstract

With the rapid advancement of global digitalization, users from different countries increasingly rely on social media for information exchange. In this context, multilingual multi-label emotion detection has emerged as a critical research area. This study addresses SemEval-2025 Task 11: Bridging the Gap in Text-Based Emotion Detection. Our paper focuses on two sub-tracks of this task: (1) Track A: Multi-label emotion detection, and (2) Track B: Emotion intensity. To tackle multilingual challenges, we leverage pre-trained multilingual models and focus on two architectures: (1) a fine-tuned BERT-based classification model and (2) an instruction-tuned generative LLM. Additionally, we propose two methods for handling multi-label classification: the base method, which maps an input directly to all its corresponding emotion labels, and the pairwise method, which models the relationship between the input text and each emotion category individually. Experimental results demonstrate the strong generalization ability of our approach in multilingual emotion recognition. In Track A, our method achieved Top 4 performance across 10 languages, ranking 1st in Hindi. In Track B, our approach also secured Top 5 performance in 7 languages, highlighting its simplicity and effectiveness\footnote{Our code is available at https://github.com/yingjie7/mlingual_multilabel_emo_detection.

JNLP at SemEval-2025 Task 11: Cross-Lingual Multi-Label Emotion Detection Using Generative Models

TL;DR

This work tackles cross-lingual, multilingual multi-label emotion detection for SemEval-2025 Task 11 by integrating a fine-tuned BERT-based classifier with instruction-tuned generative LLMs. It investigates two multi-label strategies, base (predict all labels) and pairwise (one-label-at-a-time), and examines both Track A (emotion detection) and Track B (emotion intensity) across multiple languages, including low-resource ones. Mixed-language training and the choice between BERT-based and LLM-based architectures are analyzed, with findings showing when each approach excels and how the strategies interact with label distribution and data sparsity. The approach achieves strong multilingual generalization, securing top placements in several languages and providing open-source code for reproducibility, which has practical implications for real-world cross-lingual emotion analysis in social media.

Abstract

With the rapid advancement of global digitalization, users from different countries increasingly rely on social media for information exchange. In this context, multilingual multi-label emotion detection has emerged as a critical research area. This study addresses SemEval-2025 Task 11: Bridging the Gap in Text-Based Emotion Detection. Our paper focuses on two sub-tracks of this task: (1) Track A: Multi-label emotion detection, and (2) Track B: Emotion intensity. To tackle multilingual challenges, we leverage pre-trained multilingual models and focus on two architectures: (1) a fine-tuned BERT-based classification model and (2) an instruction-tuned generative LLM. Additionally, we propose two methods for handling multi-label classification: the base method, which maps an input directly to all its corresponding emotion labels, and the pairwise method, which models the relationship between the input text and each emotion category individually. Experimental results demonstrate the strong generalization ability of our approach in multilingual emotion recognition. In Track A, our method achieved Top 4 performance across 10 languages, ranking 1st in Hindi. In Track B, our approach also secured Top 5 performance in 7 languages, highlighting its simplicity and effectiveness\footnote{Our code is available at https://github.com/yingjie7/mlingual_multilabel_emo_detection.
Paper Structure (23 sections, 4 equations, 3 figures, 6 tables)

This paper contains 23 sections, 4 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Background
  3. System Description
  4. System Overview
  5. Methods
  6. BERT-based method.
  7. LLM-based method.
  8. Experimental Setup
  9. Dataset.
  10. Evaluation Metric.
  11. Experimental Environments.
  12. Results
  13. Track A: Multi-label Emotion Detection.
  14. Development Result.
  15. Test Result.
  16. ...and 8 more sections

Figures (3)

  • Figure 1: Distribution of improved samples between strategies base and pairwise with respect to the number of emotions (track A).
  • Figure 2: Distribution of improved samples between strategies base and pairwise with respect to the number of emotions (track B).
  • Figure 3: Overall performance of the pairwise strategy across all emotional labels in all languages (Track B).