The invention discloses a social-media short text filtering method based on structure and text information. The method includes the following steps that 1, the structural characteristics of a short text are judged, and junk information is deleted; 2, the core of the text is extracted, a judge structure judges whether a retained segment text contains the core information of a described event, if nocore information exists, the information is determined as junk information, and if the core information exists, core components are extracted; 3, textual features are extracted, and the core components of the text obtained in the step 2 are mapped to a characteristic space. By scanning a participle set of the text, such structural characteristics whether junk information exists or not can be judged, and mass data in the social network is thus easily and efficiently processed; by identifying characteristics of words, sentence patterns and the like, the feature selection purpose can be achieved, based on the method in which word2vec word vectors are added so as to obtain the average, a sentence vector is constructed, the calculation amount of a classifier model in the training process is reduced, and the semantic information of the text can be well represented.