Evaluating the impact of explainable AI on user experience in a web-based natural language product recommendation chatbot

Abstract

Modern recommender systems often operate as a “Black-box”, leaving users confused or suspicious of AI recommendations. Explainable AI bridges the gap between complex recommendation algorithms and human understanding by providing explanations for its recommendations. Explainability in an industrial product recommender is essential for building user-centred AI in high-stakes settings where product features are highly specialised and mistakes can be costly. This thesis investigates the impact of providing explanations in a web-based natural language industrial product recommendation chatbot on user trust, satisfaction, and decision-making confidence. The study compares two versions of a chatbot among 42 participants, one that offers a natural language explanation and the other that does not. Each participant interacted with one version and completed a survey, which was distributed equally. The results show that users who received explanations felt more confident in the chatbot’s recommendations, reported higher trust in the system, and were more satisfied with the interaction. In contrast, users who did not receive explanations were more neutral in their trust and satisfaction, and many indicated they would have preferred to see an explanation. Participants clearly favoured the chatbot that explained its recommendations. These findings suggest that integrating explainable AI into product recommendation chatbots can lead to a better user experience by increasing user trust, satisfaction, and confidence in decision-making.