Fecha de publicación: 21/09/2016 Fuente: Wipo "BigData" The invention relates to the field of processing of natural languages, and in particular relates to a new enterprise name finding method based on a bidirectional recurrent neural network. According to the method, the bidirectional recurrent neural network is trained by automatically marking samples on existing enterprise name data, the enterprise names in a text to be processed are automatically recognized by the learning of the bidirectional recurrent neural network, and the new enterprise names not belonging to the existing enterprise names are extracted out; by automatically learning the features of basic elements of the text, such as characters, words and punctuation marks, the defect that in the traditional mode the features need to be set manually is overcome; furthermore, bidirectional propagating PNN is used in the method, so that the classification judgment result of a natural language sequence to be recognized depends on contextual information, and the accuracy rates of extraction and judgment are higher; the method provided by the invention utilizes the existing data features to find the new enterprise names, thus having very important application value in the big data analysis field, particularly the data analysis field with enterprise as the main body.