Understanding user experience in generative AI application: Evidence from topic modelling and sentiment analysis of user reviews

Abstract

Introduction. With the rapid surge of users in generative AI (GenAI) applications, user experience research is moving beyond surveys and experiments toward large-scale analyses of online reviews. Integrating topic and sentiment patterns from these reviews with user experience theory enables a clearer view of users’ concer ns and experience gaps. Method. This study analysed Chinese user reviews of DeepSeek using BERTopic for topic modelling combined with sentiment analysis and mapped the findings onto Jesse James Garrett’s five elements of user experience to reveal layered perceptions. Results. Thirty-four topics were identified and grouped into nine categories covering technical stability, device adaptation, AI interaction and functional features. Negative sentiments centred on server instability, system errors and functional deficiencies, where as positive sentiments highlighted AI performance, emotional support, and distinctive functions. The five-element mapping revealed strategic instability, scope plane functional gaps, structural interaction barriers and framework plane issues with multi-device adaptation. Conclusion. Joint topic –sentiment analysis not only uncovers core concerns of GenAI application users but also offers actionable insights for improving technical stability, device adaptation and interaction design, providing a new empirical view for optimising similar applications and advancing user experience research.