Enhancing stance detection through sequential weighted multi-task learning

The exponential growth of user-generated content on social media platforms, online news outlets, and digital communication has necessitated the development of automated tools for analyzing opinions and attitudes expressed in text. Stance detection, a critical task in Natural Language Processing, aims to identify the underlying perspective or viewpoint of an individual or group toward a specific topic or target. This paper explores the challenges of stance detection, particularly in the context of social media, where brevity, informality, and limited contextual information prevail. While sentiment analysis focuses on explicit sentiment polarity, stance detection classifies the stance or viewpoint of a text toward a target, often of an abstract nature. Motivated by recent achievements in Multi-Task Learning (MTL), this paper addresses the identified gap in the field, advocating further exploration in developing a joint neural architecture that integrates different opinion dimensions. In response, this study introduces two MTL models, Parallel Multi-Task Learning (PMTL) and Sequential Multi-Task Learning (SMTL), which incorporate sentiment analysis and sarcasm detection tasks to enhance stance detection performance. We address the complexities of MTL implementation with Transformer-based architectures and present an accessible architecture for this purpose. This study also proposes and evaluates four task weighting techniques, providing empirical evidence for their effectiveness in MTL models. Through comprehensive evaluations on benchmark datasets in both English and Arabic, we demonstrate that our most proficient model, a multi-target sequential MTL model with hierarchical weighting (SMTL-HW), achieves state-of-the-art results. These contributions underscore the potential of MTL in enhancing stance detection and offer valuable insights into the interaction between sentiment, stance, and sarcasm in text analysis.