The present invention discloses a method for extracting keywords based on the K-MEANS and the WORD2VEC. The method is characterized in that: the global semantics and the theme of each branch are integrated; a space vector is constructed by using the WORD2VEC algorithm; the K-means algorithm is used to remove the fuzzy words; the centroids distance is calculated; high quality keywords are obtained after the clustering evaluation; and by lifting the weight value, the dynamic optimization of the lexicon is realized, so that the keyword extraction has the learning evolution ability. According to the method disclosed by the present invention, the extracted keywords can reflect the internal classification theme of the document, and each keyword can well reflect the classification; and the method has the characteristics of high quality of the final keywords, wider adaptability, a more accurate result, and the like.