When a knowledge graph (KG) is growing e.g.\ by knowledge graph completion, it might become inconsistent with the logical theory which formalizing the schema of the KG. A common approach to restoring consistency is removing a minimal set of triples from the KG, called a diagnosis of the KB. However, there can be a large number of diagnoses. It is hard to manually select the best one among these diagnoses to restore consistency. To alleviate the selection burden, this paper studies automatic methods for ranking diagnoses so that people can merely focus on top diagnoses when seeking the best one. An approach to ranking diagnoses through representation learning aka knowledge graph embedding is proposed. Given a set of diagnoses, the approach first learns the embedding of the complement set of the union of all diagnoses, then for every diagnosis, incrementally learns an embedding of the complement set of the diagnosis and employs the embedding to estimate the removal cost of the diagnosis, and finally ranks diagnoses by removal costs. To evaluate the approach, four knowledge graphs with logical theories are constructed from the four great classical masterpieces of Chinese literature. Experimental results on these datasets show that the proposed approach is significantly more effective than classical random methods in ranking the best diagnoses at top places. See readme.docx in the tool package for more details.