Abstract: Ontology modularization is the key technology of ontology fusion.Some of the existing modular methods only consider the concept of limited relations in ontology,and do not consider other logical relations.Some are difficult to find the definition on internal and foreign concepts.The traditional modular approach regards all predicates as the same class,ignoring the difference in the influence of the predicate verbs on the subject.In this paper,the predicates in the ontology tuple are divided into class A predicates and class B predicates.The subject and object in the triplet are used as nodes and the predicates are used as edges,which generates the ontology structure graph.Then the weights of the edges are defined.The ontology modular problem are transformed into clustering problem,and then combining with the CH (Calinski Harabasz) clustering evaluation function to propose a modular algorithm based on the ontology structure.The method not only takes into account the influence of the predicate relation on ontology,but also transforms the vague modular classification into a clear clustering problem.The boundaries are defined more clearly,making the modular results more accurate.Finally,this method is used to divide the Pizza ontology module,which proves the validity and accuracy of the algorithm.