In earthquake-prone regions, assessing soil liquefaction potential is indispensable for contemporary seismic design. Various procedures for liquefaction triggering analysis have emerged over the past decades. However, most of them are derived from generic liquefaction databases, such that the model uncertainties in liquefaction potential assessments applied to a specific region of concern remain unknown, which poses a challenge for engineers to evaluate the liquefaction risks of target sites. This study aims to propose a hierarchical Bayesian model (HBM) to learn the inter-region characteristics of model uncertainties of the traditional simplified liquefaction potential evaluation methods based on a database containing global case histories of liquefaction categorized into several regions where those triggering events occurred. The learning outcomes can yield the model uncertainty of the target region, and the liquefaction probability at the target site under a given ground motion condition. For an illustration of the proposed model, a case history of liquefaction from a specific region is adopted to construct a quasi-region-specific model uncertainty and evaluate the liquefaction probability in the target soil. The illustration shows that the constructed quasi-region-specific model uncertainty with liquefaction histories in the target region can improve liquefaction occurrence prediction in comparison with the prediction without any histories, which is believed to benefit the engineering practice.