The problem of semantic matching is one of the core problems in natural language proces- sing. Semantic based matching, that is, calculating the degree of matching by extracting the intrinsic semantics of the text, is a hot topic in the field of natural language processing. The traditional semantic matching model does not combine sentence smoothness and other factors to perform comprehensive eva- luation, so the effect is poor. This paper proposes an enhanced semantic matching model. Based on the text similarity calculation, the model adds a smoothness factor and adjusts the optimal parameters through a large amount of data. Through the test of the automatic marking system, three commonly used automatic marking models are compared to verify that the proposed model can effectively reduce the average error value.