Expository is the most common form in text classification. With the continuous increase of information, how to improve the efficiency of expository text classification is a hotspot in researching. Keyword extraction is used to improve the efficiency in text classification. This paper suggests to extraction keywords in the text based on the use of K-means algorithm by constructing the Vector Space Model (VSM) subject to pre-assumed initial keyword frequency. The KNN text classification algorithm verifies keywords extracting based on the use of K-means algorithm not only makes the classification accuracy slightly increased, but also reduce the dimension, thus realizing the text classification efficiency.