A knowledge base dynamic screening and matching system based on an intention recognition classification model

CN122240599APending Publication Date: 2026-06-19ANSTEEL ENG TECH CORP

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: ANSTEEL ENG TECH CORP
Filing Date: 2026-04-13
Publication Date: 2026-06-19

AI Technical Summary

Technical Problem

Existing technologies suffer from problems in intent recognition and knowledge base matching, such as rigid document segmentation strategies, static hybrid retrieval weights, insufficient performance of intent recognition models, poor synergy between intent and retrieval, and a single dimension for optimizing retrieval results, resulting in insufficient accuracy and adaptability.

Method used

A dynamic filtering and matching system based on an intent recognition classification model is adopted, including a data processing module, a retrieval and matching module, and a control and optimization module. Through adaptive block segmentation strategy, hybrid retrieval and dynamic weight adjustment, multi-dimensional feature input and closed-loop optimization mechanism, a balance between high recall and high precision is achieved.

Benefits of technology

It improves the accuracy of intent recognition and knowledge base matching, especially significantly improving recall in complex reasoning scenarios, ensuring continuous optimization and adaptability of system performance, and is suitable for high-concurrency, high-precision question-answering scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122240599A_ABST

Patent Text Reader

Abstract

This invention belongs to the field of knowledge base management technology, specifically a dynamic filtering and matching system for knowledge bases based on an intent recognition classification model. It comprises three main modules: data processing, retrieval matching, and control optimization, forming a closed-loop mechanism of preprocessing, accurate retrieval, and dynamic optimization. Driven by an optimized BERT intent recognition model, this invention constructs a three-level intent recognition architecture, enabling multi-dimensional feature input and quantitative calculation of confidence metrics. It also combines the bge-m3 model to achieve adaptive semantic density segmentation, integrates Faiss vector retrieval and keyword retrieval with dynamic weighted hybrid retrieval, and utilizes intent-driven dynamic routing and continuous iteration through multi-dimensional result optimization and closed-loop feedback. This significantly improves intent recognition accuracy, accuracy in answering complex scenarios, and enhances system adaptability, accuracy, and dynamic optimization capabilities, thus meeting the high-precision question-answering needs of professional fields.

Need to check novelty before this filing date? Find Prior Art