A Multi-Stage Semantic Risk Identification Method and System Based on a Large Model

CN122090840APending Publication Date: 2026-05-26BEIJING INST OF TECH +1

View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: BEIJING INST OF TECH
Filing Date: 2026-02-09
Publication Date: 2026-05-26

AI Technical Summary

Technical Problem

Existing technologies struggle to balance security and normal user communication needs. Keyword rule-based methods have weak generalization capabilities, traditional machine learning models lack the ability to recognize complex semantic structures, and they lack interpretability and tolerance for differences in risk assessment, resulting in a high false positive rate.

Method used

We adopt a multi-stage semantic risk identification method based on a large model. Through coarse-grained risk screening, content semantic parsing, user dimension information analysis and dual-sensitivity prediction, we dynamically optimize the risk judgment threshold and combine it with a large language model for deep semantic understanding and personalized decision-making.

Benefits of technology

It achieves efficient identification of high-risk content, reduces interference with users' normal communication, enhances the ability to capture hidden and complex risks, and achieves a personalized balance between security protection and communication experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122090840A_ABST

Patent Text Reader

Abstract

This invention discloses a multi-stage semantic risk identification method and system based on a large model, relating to the field of data processing technology. The method includes: acquiring the basic dimensional features of a target user's telephone voicemail; performing a risk assessment on the basic dimensional features and outputting a first risk coefficient; if the first risk coefficient is greater than a first risk threshold, parsing the telephone voicemail content and outputting the content parsing result; acquiring the target user's prediction false positive sensitivity coefficient and prediction false negative sensitivity coefficient; optimizing and correcting the initial risk judgment threshold to generate an adaptive risk judgment threshold; activating the intent risk identification plugin embedded in the content intent recognition mechanism, performing risk identification on the content parsing result to obtain a second risk coefficient; if the second risk coefficient is greater than the adaptive risk judgment threshold, then blocking the telephone voicemail. This invention achieves an optimal balance between security protection and user communication experience at the personalized level.

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