In a specific domain, the attributes or mutual relationships connected with realistic objects are often variable dynamically and especially partially. For multi-objective categorization [2], it has been a research hotspot how to rapidly update the categorized result with the partial variations. Aiming at this problem, we propose an approach for dynamic multi-objective categorization based on the game theory, partially updating the influenced part, measuring the object's satisfaction degree by Shapley value, and improving the efficiency of calculating Shapley value by the concept of priority groups. The experimental results and the performance analyses show its better efficiency and effectiveness.