The invention relates to a minimally invasive medical field-oriented text generation method and device. In the method, vector transformation is conducted to a text, so a semantic vector and a word order vector can be obtained; an average sum of the semantic vector and the word order vector is input into an encoder to remove repetition and be simplified, so a feature vector containing important information of the text can be achieved; the feature vector is input into the encoder; a primary abstract vector is generated according to each element of the feature vector and corresponding accuracy; at last, a comparison is made between the primary abstract vector and a manual abstract vector; and when a comparison result passes, the primary abstract vector (Y) is designated as a medical history text abstract and output. Compared with a generation method in the prior art, important features can be included into the text, word repetition in a long sentence can be avoided, and abstract generation flexibility can be optimized.