This paper proposes MUSAsent, the first preference-disaggregation model to embed aspect-based sentiment directly into the linear programming structure of the MUSA method. By incorporating textual polarity signals as constraints that guide the estimation of partial satisfaction functions, MUSAsent enables a unified treatment of ratings and unstructured feedback while preserving the additive and monotonic properties of MUSA. We examine two variants in which Model I adjusts only the partial satisfaction functions and Model II extends these adjustments to the global function as well. Using thirteen real-world datasets spanning airlines, airports, lounges, hotels, online education, and beverage reviews, the study shows that Model I consistently increases post-optimality stability without degrading model fit. These findings highlight the methodological value of integrating sentiment within the core estimation process and demonstrate the practical benefit of obtaining more stable and interpretable satisfaction assessments for managerial decision-making. • MUSAsent embeds aspect sentiment into MUSA’s preference-disaggregation model. • Model I improves stability without loss of fit across 13 real-world datasets. • Unified ratings and reviews yield interpretable weights, indices, and diagrams. • Airline case shows actionable insights for service quality optimization. • Sentiment-aware action diagrams yield decision drivers for managers.
Kyriakidis et al. (Sun,) studied this question.