Emotion analysis in the development of chatbot user experience

Abstract

This thesis focuses on understanding how emotion analysis can be used to develop the user experience of digital services, such as chatbots, specifically through facial expression analysis. Using iMotions software, the study analyses emotional responses from three respondents as they interact with the service chatbot "Emma" from insurance company IF. The thesis was created as a side project to a study at Haaga-Helia University of Applied Sciences called AI Driver project. The chatbot was selected to present an example of an AI service chatbot and insurance company IF was not participating in the research. The findings reveal that chatbots, while a valuable tool in customer service, still face challenges in providing a seamless, emotionally positive user experience. Certain interactions with the chatbot, such as unclear responses or unexpected changes in the interface, generated negative emotions while successful interactions and human-like features led to positive emotions such as joy. It is evident that emotion analysis provides valuable insights into the emotional highs and lows of the interaction, making emotional data and research a significant part of designing user experiences. Additionally, this study explored the possible differences between the emotions detected by AI-based facial expressions analysis and the self-reported emotions from respondents. The research also investigated how the SERVQUAL model can be adapted to measure service quality in chatbot interaction, adding to the insight into the user experience of AI-driven customer service. Some contradicting results were found between different analysis methods which led to the conclusion that mixing several analysis tools and methods can be beneficial when analysing emotions in the user experience. As chatbots are becoming key components of customer engagement and brand interaction, the emotional highs and lows observed in the study point to a need for chatbots to not only respond accurately but also interact in a way that feels natural and human-like as well as bring joy to the user.