<p>Ordinal Priority Approach (OPA) is an optimization-based multi-attribute decision-making method that determines the weights of experts, attributes, and alternatives from ordinal rankings. To clarify its underlying structure, I first develop an equivalent formulation of OPA by mapping alternatives to rank indices. The resulting constraints explicitly involve rank order centroid (ROC) weights, showing that OPA can be interpreted as maximizing a weight disparity scalar in a ROC-weighted normalized decision space. Building on this equivalence, I then propose Generalized Ordinal Priority Approach (GOPA), an estimate-then-optimize bilevel framework that replaces fixed ROC utilities with utilities learned from partial preference information. At the lower level, I instantiate GOPA with minimum relative utility entropy (GOPA-RUE), in which expert-attribute marginal utility densities are elicited by combining a reference utility density (e.g., risk-aversion or S-shaped structures) with moment-type deterministic comparison constraints. Under this formulation, the optimal density admits an exponential-tilting, breakpoint-induced piecewise form and can be computed through a tractable linear-system reformulation, while regularized (nonnegative) least-squares corrections are used to address elicitation inconsistencies. At the upper level, once utilities have been elicited, the decision weights and the optimal disparity scalar admit closed-form solutions, yielding an efficient algorithm. Based on these results, I further establish the decomposability of the optimal weights and provide sensitivity analyses with respect to constraint, ranking, and utility perturbations. Finally, a Zhengzhou 7.20 emergency supplier selection study, together with ablation and perturbation experiments, demonstrates the flexibility and robustness of GOPA-RUE under limited and noisy preference information.</p>