Sentiment analysis technique for the development of new improved chatbot: new user humanity-like chatbot (NURUL - C)

Abstract

Emerging technologies increasingly simplify daily human tasks, with chatbots—powered by Artificial Intelligence (AI) and Natural Language Processing (NLP)—being among the most common. Although designed to simulate human conversation, chatbots often fail to meet user needs due to limitations in training data, leading to misunderstandings. This research addresses four gaps: identifying essential chatbot quality attributes, developing a text analyser, implementing sentiment analysis, and validating user satisfaction with a newly developed chatbot. Grounded in information-seeking behaviour theory and focused on the banking domain, the study followed three phases: pre-development; development, integration, and implementation; and evaluation and testing, using an iterative approach. In the pre-development phase, a pilot study with 32 customers helped refine the key quality attributes. Six were identified—functionality, efficiency, technical satisfaction, humanity, effectiveness, and ethics—with two additional attributes added later: visual elements and energy efficiency. These informed the development of the improved chatbot, NURUL-C. The system incorporated text analysis techniques such as stemming and lemmatization. In final evaluations, NURUL-C received positive feedback, supported by favourable sentiment analysis and high user satisfaction. The inclusion of energy efficiency and visual elements contributed to these outcomes. A major contribution of this work is the Holistic Evolutionary Iterative Humanistic Chatbot Development Life Cycle (HEI-HC-LC), which offers significant implications for service application development in the banking sector. Two (2) research hypotheses were tested. The first null hypothesis, “There is no difference between quality attributes of a good chatbot and customer satisfaction levels”, was rejected because there was a strong and significant correlation between chatbot attributes and customer satisfaction (r = .7689, p = .0093). The second null hypothesis (H02), “There is no relationship between positive sentiment analysis and customer satisfaction levels”, could not be rejected due to a weak and non-significant positive correlation between positive sentiment and customer satisfaction (r = .274, p = .443). Overall, users were satisfied with the new chatbot, NURUL-C.