The application discloses a method and device for model training, and a method and device for statement emotion recognition. The method for the model training comprises the steps that text information with emotion marks is acquired, wherein the text information comprises multiple statements which are marked according to pre-defined emotion types, and each statement carries the corresponding emotion mark; a statement vector corresponding to each statement in the text information is determined, wherein the statement vector is a multi-dimensional array used to represent the corresponding statement uniquely; and the emotion mark corresponding to each statement as well as the corresponding statement vector are input into a cyclic neural network, and a neural network model is obtained through the training, wherein the neural network model is used to recognize the emotion type in the statement. According to the application, the technical problem in the prior art that the recognition accuracy of emotions contained in statements is low can be solved.