This paper presents a new cascade architecture to calculate the similarity between concepts in the ontology. In the first stage of proposed model, we use path-based methods to calculate the concept of path score in the ontology. In the second stage, we use Information Content (IC)-based methods to obtain similarity scores of two extended concepts with further consideration of the two concepts of public parent set and public subset. In the third stage, we adopt a feature integration strategy to combine all the similarity scores derived from the ontology to construct various kinds of features to characterize each concept pair and using weights to balance the concept of first-stage-scores with the third-stage-scores. In the end, BP neural network is used to obtain the similarity of two concepts. This model has been evaluated and compared with existing methods when applied to the task of semantic similarity estimation. Experimental results show that the proposed method effectively improves the accuracy and scientificity of similar calculation.