Paper Type
Complete
Paper Number
PACIS2025-1513
Description
Artificial intelligence (AI) technologies are increasingly being deployed to deliver timely and personalized customer service responses. However, the inherent limitations of current AI capabilities, coupled with escalating customer demands, often result in suboptimal user satisfaction levels. This study addresses this critical challenge by investigating predictive models for post-interaction user satisfaction assessment in AI conversational services. We propose a novel framework to predict overall service satisfaction, conducting comprehensive benchmarking against established baselines. Extensive empirical evaluations on a large-scale dataset demonstrate the superior performance of our proposed model, with three key components contributing to its enhanced predictive accuracy: (1) emotional and cognitive representations that improve user state dynamics interpretation, (2) an auxiliary task predicting dissatisfaction reasons to enhance representations learning of unsatisfactory AI services, and (3) a topic contribution weighting mechanism for multiple conversational topics. These findings provide significant theoretical and practical insights for advancing AI-driven customer service interactions.
Recommended Citation
Xiao, Huiyu; Zhu, Tianyu; Wang, Lingli; and Xu, Wei, "Modeling Cognitive and Emotional Dynamics for AI-Based Conversational Satisfaction Prediction" (2025). PACIS 2025 Proceedings. 13.
https://aisel.aisnet.org/pacis2025/aiandml/aiandml/13
Modeling Cognitive and Emotional Dynamics for AI-Based Conversational Satisfaction Prediction
Artificial intelligence (AI) technologies are increasingly being deployed to deliver timely and personalized customer service responses. However, the inherent limitations of current AI capabilities, coupled with escalating customer demands, often result in suboptimal user satisfaction levels. This study addresses this critical challenge by investigating predictive models for post-interaction user satisfaction assessment in AI conversational services. We propose a novel framework to predict overall service satisfaction, conducting comprehensive benchmarking against established baselines. Extensive empirical evaluations on a large-scale dataset demonstrate the superior performance of our proposed model, with three key components contributing to its enhanced predictive accuracy: (1) emotional and cognitive representations that improve user state dynamics interpretation, (2) an auxiliary task predicting dissatisfaction reasons to enhance representations learning of unsatisfactory AI services, and (3) a topic contribution weighting mechanism for multiple conversational topics. These findings provide significant theoretical and practical insights for advancing AI-driven customer service interactions.
Comments
AI ML