Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detecting Hope Across Languages: Multiclass Classification for Positive Online Discourse

T. O. Abiola, K. D. Abiodun, O. E. Olumide, O. O. Adebanji, O. Hiram Calvo, Grigori Sidorov

TL;DR

The paper tackles multilingual multiclass detection of hopeful online discourse across English, Spanish, German, and Urdu by fine-tuning XLM-RoBERTa with a multilabel head and integrating active learning to handle class imbalance. It defines four classes (Not Hope, Generalized, Realistic, Unrealistic Hope) and evaluates on the PolyHope dataset, demonstrating that transformer-based multilingual models outperform logistic regression baselines and achieve competitive macro F1 scores across languages, including low-resource Urdu. Key contributions include a comprehensive methodology combining dataset distribution awareness, weighted loss, and iterative uncertainty-driven data selection, along with a comparative analysis against existing techniques. The work advances practical positive-content moderation by enabling robust, cross-lingual hope speech detection with fine-grained categorization, while acknowledging limitations in data diversity and computation that future work can address.

Abstract

The detection of hopeful speech in social media has emerged as a critical task for promoting positive discourse and well-being. In this paper, we present a machine learning approach to multiclass hope speech detection across multiple languages, including English, Urdu, and Spanish. We leverage transformer-based models, specifically XLM-RoBERTa, to detect and categorize hope speech into three distinct classes: Generalized Hope, Realistic Hope, and Unrealistic Hope. Our proposed methodology is evaluated on the PolyHope dataset for the PolyHope-M 2025 shared task, achieving competitive performance across all languages. We compare our results with existing models, demonstrating that our approach significantly outperforms prior state-of-the-art techniques in terms of macro F1 scores. We also discuss the challenges in detecting hope speech in low-resource languages and the potential for improving generalization. This work contributes to the development of multilingual, fine-grained hope speech detection models, which can be applied to enhance positive content moderation and foster supportive online communities.

Detecting Hope Across Languages: Multiclass Classification for Positive Online Discourse

TL;DR

The paper tackles multilingual multiclass detection of hopeful online discourse across English, Spanish, German, and Urdu by fine-tuning XLM-RoBERTa with a multilabel head and integrating active learning to handle class imbalance. It defines four classes (Not Hope, Generalized, Realistic, Unrealistic Hope) and evaluates on the PolyHope dataset, demonstrating that transformer-based multilingual models outperform logistic regression baselines and achieve competitive macro F1 scores across languages, including low-resource Urdu. Key contributions include a comprehensive methodology combining dataset distribution awareness, weighted loss, and iterative uncertainty-driven data selection, along with a comparative analysis against existing techniques. The work advances practical positive-content moderation by enabling robust, cross-lingual hope speech detection with fine-grained categorization, while acknowledging limitations in data diversity and computation that future work can address.

Abstract

The detection of hopeful speech in social media has emerged as a critical task for promoting positive discourse and well-being. In this paper, we present a machine learning approach to multiclass hope speech detection across multiple languages, including English, Urdu, and Spanish. We leverage transformer-based models, specifically XLM-RoBERTa, to detect and categorize hope speech into three distinct classes: Generalized Hope, Realistic Hope, and Unrealistic Hope. Our proposed methodology is evaluated on the PolyHope dataset for the PolyHope-M 2025 shared task, achieving competitive performance across all languages. We compare our results with existing models, demonstrating that our approach significantly outperforms prior state-of-the-art techniques in terms of macro F1 scores. We also discuss the challenges in detecting hope speech in low-resource languages and the potential for improving generalization. This work contributes to the development of multilingual, fine-grained hope speech detection models, which can be applied to enhance positive content moderation and foster supportive online communities.

Paper Structure

This paper contains 25 sections, 4 equations, 6 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Methodology
  4. Datasets
  5. Dataset Distribution
  6. Preprocessing
  7. Active Learning Strategy
  8. Uncertainty Sampling
  9. Selection and Learning
  10. Incorporating Class Weights
  11. Benefits to Learning
  12. Model Architecture
  13. Logistic Regression
  14. XLM-RoBERTa
  15. Evaluation and Prediction
  16. ...and 10 more sections

Figures (6)

  • Figure 1: Class Distribution Heatmap
  • Figure 2: Comparative line graph of Macro F1 and Micro F1
  • Figure 3: English Dev Set Confusion Matrix
  • Figure 4: German Dev Set Confusion Matrix
  • Figure 5: Spanish Dev Set Confusion Matrix
  • ...and 1 more figures