Knowledge Graphs (KGs), which contain rich relational information, have been wide-ly utilized in various tasks. However, there may exist inconsistent facts in KGs, especially in automatically constructed large-scale KGs. To address this problem, we innovatively propose an entity&paths semantics based multi-classification model to solve the problem of inconsistency detection. It synthesizes the internal semantic information both in entity and relation level of the KG to measure the association strength between triples so that different kinds of inconsistencies can be accurately detected. We conduct experiments in the real-word dataset FB15k (from Freebase) and the results show that our approach achieve significant and consistent improvement compared to existed advanced approaches, confirming the capability of our framework in knowledge graph inconsistency detection.