Enhancing Emotion Understanding in Messaging Interactions

This dissertation explores the challenges of understanding emotions in messaging conversations with task-oriented conversational agents. It presents a novel approach to address this issue using a combination of natural language processing (NLP) methods and Affect Control Theory (ACT). The lack of nonverbal cues and the ambiguous nature of written language makes understanding emotions in messaging conversations a complex task. However, the ability to understand and respond to emotions can significantly improve the effectiveness and satisfaction of communication between humans and conversational AI systems.
The research has three main components. The first phase demonstrates a proof of concept of using ACT in the context of messaging with a task-oriented conversational agent. The second phase addresses the limitations of traditional affective dictionaries used in ACT by utilizing the capabilities of BERT, a pre-trained transformer-based model. The final phase focuses on recognizing emotions during conversations with a task-oriented conversational agent by utilizing both the sequential and contextual aspects of the messages.
The results showed that combining NLP methods with ACT can provide a more accurate and complete understanding of the emotions conveyed in messaging interactions. The study also achieves state-of-the-art results in estimating emotions by implementing an encoder-decoder network with attention. This approach can lead to improved emotional intelligence in conversations with chatbots and result in enhanced customer satisfaction and more natural and effective conversational AI systems in various domains such as customer service, healthcare, and education. Overall, this research contributes to the field of conversational AI by making strides in understanding emotions in messaging interactions and applying NLP and ACT to improve conversational AI systems.