Identification of sarcasm in texts for sentimental analysis
Humanity has discovered various ways to express emotions. Depending on the context of speech, these emotions are sometimes accompanied by sarcasm, particularly when expressing intense feelings. Over the past few decades, social media platforms such as Facebook, Instagram, TikTok, Twitter, and YouTube have become popular tools for people to share such strong emotions and personal thoughts with wide audiences. Through techniques like sentiment analysis, this data can be valuable in various fields, including business, marketing, production, behavioral analysis, and public management during ecological or biological crises, as well as in times of war.
Most current research treats sentiment and sarcasm classification as two separate tasks, approaching each as an independent text classification problem. In recent years, studies using deep learning algorithms have significantly improved the effectiveness of these independent classifiers. However, one of the main challenges these approaches face is their inability to accurately classify sarcastic statements as negative. Taking this into account, we argue that recognizing sarcasm enhances sentiment classification, and vice versa. In this work, we demonstrate that these two tasks are interrelated. This paper proposes a multi-task learning framework that leverages deep neural networks to model this interrelation, aiming to improve the overall effectiveness of sentiment analysis.