The invention discloses a relation classifier based on a selectivity attention convolution neural network and a method of the relation classifier and relates to the technical field of relation extraction and classification. The relation classifier comprises a coding layer, a convolution layer, a selectivity attention layer and a pooling and softmax layer successively. The coding layer is used for converting words in sentences into low-dimension real number vectors. The convolution layer is used for acquiring high level features of each word. The selectivity attention layer is used for finding out two words with the closest entity semantic relations through a shortest dependency path and displaying the words through a weight matrix. The pooling and softmax layer is used for constructing sentence vector expressions and carrying out classification to obtain a target relation. According to the invention, by combining sentence meaning information and word meaning information, features are quite effectively extracted; and syntactic structure information is considered on the CNN and selectivity attention is added to key words, so final classification results are improved.