Method and apparatus for segmenting training samples

By segmenting the training samples into dialogue paragraphs and event dialogue fragments corresponding to the topic and establishing a correspondence with structured forms, the problem of poor training results caused by lengthy texts is solved, and the training efficiency of the model is improved.

CN115271098BActive Publication Date: 2026-03-31ALIPAY (HANGZHOU) INFORMATION TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-08-01
Publication Date
2026-03-31

AI Technical Summary

Technical Problem

In existing structured summarization model training methods, lengthy dialogue texts prevent the model from fully utilizing training samples, resulting in poor training performance.

Method used

The training samples are segmented into dialogue paragraphs corresponding to multiple topics and dialogue fragments corresponding to events, and a corresponding relationship is established with the structured form blocks to form independent training samples.

Benefits of technology

By segmenting the training samples, the problem of the model being unable to handle lengthy text was avoided, thus improving the training effect.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115271098B_ABST
    Figure CN115271098B_ABST
Patent Text Reader

Abstract

The embodiments of the present specification provide a method and device for segmenting training samples. In the method, the dialogue text as the training sample is segmented according to the theme to obtain dialogue paragraphs corresponding to each theme. The structured form as the training sample is divided into multiple form blocks according to the theme, and each form block corresponds to a theme. When the same type of event included in each theme can occur multiple times, the dialogue paragraph corresponding to the theme is segmented according to the type of event to obtain dialogue segments corresponding to each event. Each obtained dialogue segment and a sub-block in the form block corresponding to the dialogue segment are determined as a corresponding relationship. Each dialogue segment and the corresponding sub-block and each dialogue paragraph and the corresponding form block that are not segmented are determined as training samples.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Theme information-based text segmentation method

    CN110110326A

  • Medical record text data segmentation method and device, readable storage medium and electronic equipment

    CN112786132A

  • Document information extraction method and device, equipment and medium

    CN114821590A