Aspect sentiment triple extraction method for labeling packaging strategies

By employing an aspect-based sentiment triple extraction method based on a labeled packaging strategy, an entity recognition and sentiment classification model is constructed. Utilizing graph convolutional neural networks and multilayer perceptrons, the problems of inaccurate entity recognition and neglect of span relationships in existing technologies are solved, achieving more efficient triple extraction and faster inference speed.

CN117951301BActive Publication Date: 2026-07-21GUILIN UNIV OF ELECTRONIC TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
GUILIN UNIV OF ELECTRONIC TECH
Filing Date
2024-02-19
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing technologies suffer from inaccurate entity recognition, insufficient sentiment classification, and neglect of span relationships in triple extraction, resulting in poor triple extraction performance.

Method used

A tagging and packing strategy is adopted. By constructing an initial entity recognition model and a sentiment classification model, a graph convolutional neural network is used to fuse dependencies. A multilayer perceptron is used for sentiment classification. Type tags are inserted to emphasize subject span features. The relationship between span pairs is processed through a neighborhood-oriented packing method and the parallelism of dangling tags.

Benefits of technology

It improves the accuracy and speed of triple extraction, especially when dealing with complex relationships and multi-word spans, enhances the model's ability to identify entity boundaries and types, and improves F1 score and inference speed.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117951301B_ABST
    Figure CN117951301B_ABST
Patent Text Reader

Abstract

The application discloses an aspect sentiment triple extraction method of a marking packaging strategy, and comprises the following steps: constructing an initial entity recognition model and an initial sentiment classification model, independently training the initial entity recognition model and the initial sentiment classification model, and obtaining an entity recognition model and a sentiment classification model; inputting a to-be-tested sentence into the entity recognition model, obtaining aspect words and viewpoint words in the sentence, and constructing an input of the sentiment classification model based on the entity predicted by the entity recognition model; putting the input of the sentiment classification model into a pre-trained language model, obtaining a feature vector of each word and a label, inputting the feature vector into a graph convolutional neural network, fusing a dependency relationship in the sentence, obtaining a final feature vector, performing sentiment classification on the feature vector through a multilayer perception machine, and obtaining an aspect sentiment triple. The application can improve the triple extraction effect.
Need to check novelty before this filing date? Find Prior Art