Predicate central word recognition method based on neural network

The invention discloses a predicate center word recognition method based on a neural network. The method comprises the following steps of 1, performing vector mapping on a text based on a pre-trainingword vector and a random word vector; 2, acquiring features and long-term dependency relationships of sentences through a neural network model; 3, relieving the problem of gradient disappearance in the depth model by using a Highway network; and 4, constraining the output path marked by the sequence through a constraint function. According to the method, a long-term dependency relationship in a sentence is obtained through multi-layer Bi-LSTM superposition, then the problem of gradient disappearance of the deep layer model is relieved through highway connection, finally, normalizing is performed through a Softmax layer to obtain a maximum annotation path, and in addition, an output path is planned through a constraint function to solve the uniqueness problem of a predicate center word. 本发明公开了一种基于神经网络的谓语中心词识别方法，所述