Feature Selection (FS) is reducing dimensions and denoising. However, there are many factors that affect the features selection, mainly including the dimensions, importance, and semantic of terms. For feature representing high-dimensional but sparse of short text and traditional features extraction lack semantic, a feature selection function FS fusing multi-factors is constructed. It is verified that FS not only can integrate the semantics of the feature, but also can remove a large number of redundant features, thus improve the weight of the features with class distinction capabilities, comparing with the traditional feature selection function TF-IDF. FS as a new function, using the Chinese corpus of Sogou Lab for short text classification, verifys the effectiveness of the method.