Neutral text generation method based on pre-classifier
By using a pre-classifier-based approach combined with BERT and matrix factorization techniques, a text manipulation strategy is determined, and neutral text is generated using the BART model. This solves the accuracy problem of end-to-end models in eliminating subjective bias in text generation and achieves a significant improvement in neutral text generation.
CN117390189BActive Publication Date: 2026-04-07BEIJING UNIV OF TECH
2 Cites 0 Cited by
Patent Information
- Application Number
- CN202311458205.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-05
- Publication Date
- 2026-04-07
- Estimated Expiration
- 2043-11-05
AI Technical Summary
Technical Problem
Existing end-to-end deep learning models lack fine-grained text manipulation strategies when eliminating subjective biases in text, resulting in low accuracy in generating neutral text.
Method used
We employ a pre-classifier-based approach, using a sequence labeling model to label biased words, and combining BERT and matrix factorization techniques to extract syntactic and semantic features. This determines deletion or replacement strategies, and we use the BART model to generate neutral text.
Benefits of technology
The accuracy of generating neutral text was improved, with the end-to-end model's anti-bias accuracy increasing from 45.8% to 51.8%.
✦ Generated by Eureka AI based on patent content.
Abstract
The application discloses a neutral text generation method based on a pre-classifier. The existing method uses the same text generator to replace or delete the words in the sentence to eliminate subjective bias, however, such a method is affected by different operation distributions, when one operation is less, the model is more inclined to perform another operation. The method is characterized by fine-grained analysis and processing of the debiasing problem, and improves the debiasing effect. The method mainly includes three steps: sequence labeling model marking the operation sequence; classifier judging the operation strategy (extracting the text semantic vector; text syntax feature embedding; feature fusion; MLP realizing classification); and neutral text generation. The method combines the semantic features and syntax features of the sentence to analyze the operation type in a fine-grained manner and mark the operation sequence, and according to the operation type (replacement or deletion), the corresponding generation strategy is adopted to obtain the neutral sentence.
Need to check novelty before this filing date? Find Prior Art