Long text matching method combining noise filtering and divide-and-conquer strategy

By combining noise filtering and partitioning strategies, a long text matching model is constructed, which solves the problem of long text matching in the existing technology, and achieves high accuracy and efficient matching.

WO2025060268A1 Publication Date: 2025-03-27SHENZHEN INST OF ADVANCED TECH CHINESE ACAD OF SCI
4 Cites 6 Cited by

Patent Information

Application Number
PCT/CN2023/137232
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-09-18
Filing Date
2023-12-07
Publication Date
2025-03-27

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively solve the problem of long text matching, especially due to the rich semantics and complex logical relationships of long texts. Existing methods such as Match-Ignition and Divide and Conquer have shortcomings in noise filtering and matching granularity processing.

Method used

A long text matching method combining noise filtering and division and governance strategies is proposed. A long text matching model is constructed through keyword extraction layer, filtering layer and association extraction layer, and noise is filtered using PageRank algorithm, and the long text matching is decomposed into two sub-problems of keyword and association matching through division and governance strategy, minimizing the KL-divergence between the global matching distribution and the keyword and association matching distribution.

Benefits of technology

It significantly improves the accuracy of long text matching, can effectively process granular information at different levels of matching granularity in long text, and improves the generalization ability and matching efficiency of the model.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

Disclosed is a long text matching method combining noise filtering and a divide-and-conquer strategy. The method comprises: constructing a long text matching model, which comprises a keyword extraction layer, an association extraction layer and a filtering layer, wherein the keyword extraction layer is used for extracting keywords in texts, the filtering layer is used for filtering noise in the texts on the basis of the similarity between sentences to obtain a denoised text sequence, and the association extraction layer is used for removing the keywords from the denoised text sequence to obtain the remaining associated texts; taking minimization of a set overall loss function as an optimization target to train the long text matching model, wherein the loss function reflects global matching distribution and combines the keywords with associated matching distribution; and for a target text, using the trained long text matching model to perform real-time text matching. The present invention improves the generalization ability and accuracy of text matching.
Need to check novelty before this filing date? Find Prior Art