Multi-Task Learning for Sentiment and Topic Classification in Social Media Texts

Abstract

Social media platforms have emerged as rich sources of textual data, offering insights into public opinion, consumer preferences, and emerging topics. However, extracting meaningful information from such unstructured and noisy content presents considerable challenges. This study proposes a multi-task learning (MTL) framework to simultaneously perform sentiment classification and topic classification on social media texts. By sharing representations across tasks, the model leverages interrelated patterns and dependencies between sentiment and topical content. Experimental results demonstrate that the MTL approach outperforms single-task baselines in both accuracy and generalization, especially in scenarios with limited labeled data. The integration of attention mechanisms and transformer-based encoders further enhances model interpretability and robustness, offering a scalable solution for real-time social media analytics.