Unlock instant, AI-driven research and patent intelligence for your innovation.

An Improved Hierarchical Sequence Labeling Joint Relation Extraction Method Based on Neural Network

A layered sequence and relational extraction technology, applied to biological neural network models, neural architectures, instruments, etc., to achieve the effect of alleviating the problem of entity nesting and accurate entity extraction

Active Publication Date: 2022-05-17
TIANJIN UNIV
View PDF4 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0007] Aiming at the defects existing in the existing relationship extraction model, the present invention provides an end-to-end neural network-based improved layered sequence labeling joint relationship extraction method. The designed CNN module is connected to the pre-training model BERT to replace the traditional full The connection layer performs the sequence labeling of all the text in the whole sentence, and uses an improved "half pointer-half label" method on the basis of the traditional sequence labeling to enhance the effect of entity extraction

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • An Improved Hierarchical Sequence Labeling Joint Relation Extraction Method Based on Neural Network
  • An Improved Hierarchical Sequence Labeling Joint Relation Extraction Method Based on Neural Network
  • An Improved Hierarchical Sequence Labeling Joint Relation Extraction Method Based on Neural Network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0039] An improved hierarchical sequence annotation joint relation extraction method based on neural network, see figure 2 , the method includes the following steps:

[0040] 101: Input the text into the model, and obtain the text feature vector output through the pre-training model BERT.

[0041] 102: Decode the text feature vector through a convolutional neural network module (Convolutional Neural Networks, CNN), and output the subject's head position mark sequence.

[0042] It can be noticed that the present invention uses a well-designed CNN module to perform sequence labeling after the pre-training model BERT. Most of the existing methods for sequence annotation models are very simple and have limited ability to fuse contextual information. This method effectively utilizes the advantage that the CNN module pays more attention to local information, and can also supplement more location features to a certain extent, so as to perform more accurate annotation.

[0043] 10...

Embodiment 2

[0049] The scheme in embodiment 1 is further introduced below in conjunction with specific examples and calculation formulas, see the following description for details:

[0050] 201: Input the text into the model, and obtain the text feature vector through the pre-training model BERT.

[0051] Among them, the above step 201 mainly includes: preprocessing the input text, truncating or supplementing it according to the specified length n, inputting BERT, and according to the BERT word list, each word has its corresponding ID, so the corresponding ID of the text sequence can be obtained. ID sequence, the length is n. Then input the ID sequence into the BERT model to obtain the output text feature vector where n represents the length of the text and k represents the dimension of the text feature vector for each word.

[0052] 202: Convert the text feature vector Decoding by a CNN module outputs a sequence of head position markers for the subject.

[0053] Wherein the CNN mod...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The present invention relates to an improved layered sequence annotation joint relation extraction method based on neural network, comprising: inputting text into a model, obtaining a text feature vector through a pre-training model; decoding the text feature vector through a CNN module, and outputting the head position of a subject Mark sequence; fuse the head position mark sequence of the subject with the text feature vector, decode through the CNN module, and output the tail position mark sequence of the subject; fuse the subject's prior information with the text feature vector to form a new text feature vector, pass the CNN module Decode, output the head position mark sequence of the object corresponding to the subject's ownership relationship; then fuse the head position mark sequence of the object with the text feature vector to form a new text feature vector, decode through the CNN module, and output the object under the corresponding subject's ownership relationship The tail position tag sequence of , and complete the decoding of the relationship and object at the same time; output the triples contained in the text according to the head and tail position tag sequence of the subject and object.

Description

technical field [0001] The invention relates to the technical field of natural language processing, in particular to an improved layered sequence labeling joint relation extraction method based on a neural network. Background technique [0002] As a subtask of information extraction, relational extraction is to automatically identify entities and the relationship between entities through certain technical methods in the case of given unstructured or semi-structured text. With the continuous development of relational extraction technology in the field of natural language processing, the hidden information in a large amount of structured, semi-structured, and unstructured massive data generated by various industries can be excavated and redeveloped to provide social progress and industry development. A new driving force and a guiding role in development. [0003] Early relational extraction models were mainly based on rules and templates, with high accuracy but low recall. T...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06F40/295G06F40/216G06K9/62G06N3/04
CPCG06F40/295G06F40/216G06N3/045G06F18/253
Inventor 高镇庞佳佳
Owner TIANJIN UNIV