Abstract: The problem of large feature dimension and the expansibility reduces the KNN function. This paper brougt forward an ameliorative KNN method to solve the problem that big swatch sort with more texts is easy to become the knearest neighbors under the feature reduction condition. Firstly, it used an improved odds radio method to select feature.Secondly, it estimated the possible sorts for the text by using the sort vector. Lastly, it used an improved KNN method in the reduced texts to realize text categorization. The experiment shows that this method has improved the precision.

Key words: KNN, Feature reduction, Feature selection, Text categorization