There are many questions without interrogative words in the Community Question Answering (CQA), where non-factoid questions make up a high proportion. In order to solve a specific case that the traditional categories for question classification is based on the factoid questions and the traditional methods for question classification largely depend on the interrogative words, a coarse-grained classification category and a novel hierarchical structure question classification method based on the interrogative words were proposed. The Support Vector Machine (SVM) model was used to classify the questions which contained interrogative words. As for the questions without interrogative words, a classifier based on focus word was constructed. The comparison experiment with method based on SVM was conducted on the dataset of Chinese questions crawled from Zhihu, and the proposed method improved the accuracy by 4.7 percentage points. The experimental results illustrate that the proposed method which selects different classifier according to whether a question contains interrogative words can effectively reduce the dependence on interrogative word, and make more accurate classification for Chinese questions.