A Text-Based Hybrid Approach for Multiple Emotion Detection Using Contextual and Semantic Analysis

With the growing importance of textual data processing, sentiment analysis which is a field of text mining, has been widely researched. But it is insufficient for the detection of human emotions. Emotion detection, an extension of sentiment analysis, has proven to be one of the most important areas in text mining, especially in the field of human-computer interactions. The recent works on emotion detection primarily focus on facial expressions, voice, audio and gestures. However, the content on the web is mostly text-based and it becomes difficult to capture the human emotions in the absence of facial and audio aspects in the data. Therefore, there is a need to design efficient mining techniques for processing textual data. Traditional approaches overlook disambiguation and ignore the presence of multiple emotions in text. In this paper, we propose a hybrid model which uses rules, sentiments and context for the disambiguation of words by using sentence transformers which recognize the various emotions involved by using natural language processing, sentence embeddings, BERT and similarity techniques so as to overcome such shortcomings. Our work ensures that Ekman's emotions along with neutral emotion are identified such that multiple emotions are tagged precisely based on the context. This hybrid method has proven to be far superior than existing approaches for the detection of multiple emotions.