The invention provides a semantically-enhanced classification method for music reviews, which is used for automatically classifying the music reviews. Via the method, music review contents are automatically divided into five classes, namely an album review, an artist review, a melody review, a lyric review and an external characteristic review, and automatic classification for lots of classes of unknown music reviews can be realized by only performing manual classification on a small quantity of music reviews. The method provided by the invention has the following advantages that not only are the music reviews analysed, the accuracy of the analysis is enhanced by applying music phrases related to the reviewed object, such as music metadata and lyrics; and simultaneously, via the method provided by the invention, a method for effectively combining the outputs of two classifiers is further provided, thus remarkably enhancing the accuracy of a prediction.