A multi-granularity entity relation extraction method

Abstract

The application belongs to the technical field of natural language processing, and particularly relates to a multi-granularity entity relation extraction method; the method comprises the following steps: inputting preprocessed training text into a sentence encoder for processing to obtain a sentence hidden state vector of each layer and an entity hidden state vector of two words; constructing an entity hidden state vector matrix under multiple granularities and calculating the final representation of the two words according to the matrix; performing weighted layer pooling processing on the sentence hidden state vector of each layer, splicing the final representation of the two words and the sentence sequence encoding result obtained by the pooling processing to obtain a sentence comprehensive representation; reconstructing a label and encoding the reconstructed label; performing comparative learning according to the sentence comprehensive representation and the reconstructed label encoding; calculating a ternary loss and adjusting model parameters according to the ternary loss to obtain a trained model; and the application can solve the problems of lacking of simultaneously considering global and local information of a text and lacking of transfer learning effect of obtained text encoding results.

Images