Text entity relationship prediction method based on neural network model

Abstract

The invention discloses a neural network model-based text entity relationship prediction method, which relates to the technical field of text processing, and comprises the following steps of: inputting a text into a bidirectional long short-term memory (BI-LSTM) model to obtain a plurality of entities in the text, and respectively inputting the plurality of entities into a convolutional neural network (CNN) model to obtain feature vectors corresponding to the entities; respectively inputting the words on the left sides and the words on the right sides of the two entities into a BILSTM model toobtain feature vectors fleft of the words on the left sides of the two entities and feature vectors fright of the words on the right sides of the two entities, and inputting the words between the twoentities into a CNN model to obtain feature vectors fmid of the words between the two entities, according to the method, fe1, fe2, fmid, fleft, flight and fdist are spliced into a vector to be inputinto a feedforward neural network model, and the label with the maximum probability value is used as the relationship between two entities in the text, so that the accuracy and applicability of relationship prediction between the text entities are improved.

Description

technical field [0001] The invention relates to the technical field of text processing, in particular to a method for predicting the relationship between text entities based on a neural network model. Background technique [0002] In Chinese natural language processing, entity relationship extraction refers to identifying entities in the text and extracting the relationship between named entities for a random piece of text. The so-called entity refers to the time, place, etc. that appear in the text; the so-called relationship refers to the semantic connection between entities. [0003] Most of the current entity relationship prediction methods are performed in a pipelined manner, such as inputting a sentence, first performing entity recognition, then combining entities in pairs, and then performing relationship extraction. This method has the following disadvantages: [0004] (1) The wrong choice of the entity recognition module will affect the performance of relation ext...

AI Technical Summary

Problems solved by technology

[0004] (1) The wrong choice of the entity recognition module will affect the performance of relation extraction, resulting in a high error rate;

[0005] (2) The relationship between two subtasks cannot be extracted. For example, for the text "Ren Zhengfei is employed by Huawei", Ren Zhengfei and Huawei have an employment relationship. It can be known that one entity belongs to the organization type and the latter entity belongs to the person type, but the pipeline cannot be used This information is not applicable;

[0006] (3) Redundant information will be generated. Pair the entities identified by entities in pairs, and then perform relationship extraction. Entity pairs without relationships will bring redundant information, and the error rate is high.

Images