Clustering is an important task in data mining, and fuzzy clustering is on the significant status in clustering, which can deal with all types of datasets, has been at the center of research interest in recent years. The clustering method in this paper is based on domain knowledge, from which we can obtain the tuples? semantic proximity matrix, then two clustering methods are introduced, which both started from semantic proximity matrix, so the results of clustering can be instructed by domain knowledge. The two clustering methods are Natural Method (NM) and Graph-Based Method (GBM), which are both controlled by a threshold that is confirmed by polynomial recession. Theoretical analysis testify the corrective of our approach, the extensive experiments on synthetic datasets compare the performance of our approach with that of Modified MM approach in literature [1] and highlight the benefits of our approach, and the experimental results on real datasets discover some rules which are useful to domain experts.