[Purpose/significance] This paper aims to propose a mapping model based on cosine similarity for mapping between patent documents and industrial classification. This model is accurate, efficient and scalable, which provides some references for the further research. [Method/process] After introducing the methods for mapping between patents and industrial classification, we designed a model for mapping between patent documents and industrial classification and completed the mapping between the 2015 annual patents of Chinese Academy of Sciences and the Classification of Strategic Emerging Industries. Then we evaluated this model according to the mapping results. [Result/conclusion] This model obtains the mapping results between patent documents and industrial classification automatically by the natural language processing technology and enables mapping between patents and industrial classification bi-directionally. The method saves a lot of labor costs and can easily adjust the fine-grained classification and be applied to most of the mapping between patents and industrial classification. Finally, improvements of the model are described. Some future application areas are also briefly discussed in this paper.