The short text has the characteristics of less vocabulary, more noise and sparse features, which leads to the unsatisfactory effect of the traditional text classification method applied to the short text classification. In order to improve the classification accuracy of short texts, a feature extension method based on Wikipedia word vector is proposed. First, word vectors are trained using Wikipedia corpus. Then, word vector is combined with document vector for feature selection. Finally, by extending the word set with high similarity of feature items, the resulting text is classified by the traditional classifier. Experimental results show that the proposed method is better than other text feature extension algorithms in the accuracy of short text classification.