Microservice architecture (MSA) has become the de facto standard for designing cloud-native enterprise applications due to its efficient infrastructure setup, high service availability, and elastic scalability. It is a challenging task for enterprises to decompose monolithic applications into microservice architectures. Existing microservice extraction methods overlook database table relationship, leading to excessive database table synchronization between extracted microservices and imbalanced microservice architecture performance. This paper proposes CSMO, a Cross-Supervision Method for Microservice Optimization through Decentralized Data Management, aimed at refining existing microservice extraction methods. CSMO constructs three types of interclass relationship view for monolithic applications, including the interclass call relationship view, the inter-class database table relationship view, and the inter-class business association relationship view, to guide the optimization of existing microservice extraction results. Comparative experiments have been conducted on two monolithic applications against four methods. The results demonstrate that CSMO significantly outperforms the baseline methods in all aspects, particularly reducing the frequency of database table synchronization by an average of 51% after optimization.