Classifying Emotion Using Multi-Label Deep Learning Techniques

An emotion is a complicated psychological phenomenon that encompasses various reactions occurring in the brain and body, triggered by cognitive content. A significant challenge in natural language processing is the classification of emotions in text with multiple labels to predict the range of emotions associated with a specific text that accurately represents the author’s psychological state and is applicable in various real-world data mining applications. Nonetheless, the process of annotating multiple emotions in text classification is inherently costly and labour-intensive. A considerable amount of research has been conducted on text emotion classification in both English and Chinese. However, most research has concentrated on single-label emotion classification, which fails to recognize all the emotions present in each instance. Yet, there has been extensive investigation into multi-label emotion classification. Therefore, we developed a novel approach to enhance the accuracy of emotion classification by integrating vectorization techniques with deep learning models. Initially, the text input is preprocessed and transformed into numerical vectors using methods such as term frequency–inverse document frequency or word embeddings. Following this, a deep learning architecture with multiple layers, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), processes the vectorized data. The deep learning model enables precise emotion classification by effectively identifying correlations and patterns in the text data. Experimental results on a benchmark dataset illustrate the effectiveness of the developed model in achieving high classification accuracy for emotions.