Measuring service quality and related key dimensions has been an important problem in Marketing. In this research, we would introduce a smart methodological framework to efficiently identify refined, key sentiment dimensions for measuring the service quality using both traditional survey and unstructured online reviews (natural survey). The proposed framework consists of three parts: (1) steps for preprocessing the unstructured reviews to generate attribute-level sentiments for independent variables (2) Bayesian regression to efficiently identify key groups of correlated attributes and (3) post-hoc analysis for identifying dimensions from the selected groups of correlated attributes and predicting dimension-level effects. Note, the first part of the framework (i.e., preprocessing) is not required for analyzing traditional surveys. Our framework provides two sets of complementing results such as attribute-level effects under the identified dimensions and aggregate dimension-level effects. In the first application study to traditional SERVQUAL data, we successfully validated the proposed framework by comparing the results between our framework and three commonly used existing methods of regression, lasso regression, and factor analysis. In the second empirical application study with the online reviews from a major game review website, STEAM platform, we found that our framework provided a significantly reduced number of key dimensions which were surprisingly efficient for predicting and explaining the service quality ratings, compared with the same set of compared methods in the first study plus the topic model. In particular, with reviews of 2,825 games, three key dimensions of Mechanical playability, Fun in fantasy and Money for value were identified, and we also found that the Mechanical playability could be an important driver of game popularity.