Multi-dimensional consistency detection method and system in large model

CN122021606APending Publication Date: 2026-05-12BEIJING UNIV OF POSTS & TELECOMM
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
BEIJING UNIV OF POSTS & TELECOMM
Filing Date
2025-12-19
Publication Date
2026-05-12

AI Technical Summary

Technical Problem

When generating content, large models suffer from issues such as time relevance, context dependence, outdated knowledge references, retelling of old news, misplaced context, and disordered event timelines in cross-segment reasoning tasks. These issues lead to a decline in content quality and may mislead users, especially in high-risk scenarios where they pose potential harm. Existing technologies struggle to effectively identify and correct these biases.

Method used

By constructing time-link and contextual structure information, quantifying time consistency scores and contextual consistency scores, generating a time-series risk index, and comprehensively evaluating the degree of time-series deviation output by the large model, including event extraction, timestamp mapping, context window analysis, and entity consistency assessment, a quantifiable and interpretable detection mechanism is provided.

Benefits of technology

It enables comprehensive review of the output content of large models in terms of time and context, can identify and quantify time-series deviations, provide explainable anomaly explanations and repair suggestions, improve the time-series consistency and security of content, and is suitable for high-risk scenarios such as news generation and legal consultation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122021606A_ABST
    Figure CN122021606A_ABST
Patent Text Reader

Abstract

The invention provides a multi-dimensional consistency detection method and system in a large model, and the method comprises the steps: obtaining a text generated by the large model, and processing the text to generate a sentence sequence containing time information; an event is extracted from the text, the event and time information are combined to construct a time link where the event occurs, the time link is quantified, a time consistency score is generated, and the time consistency score is used for evaluating the logic rationality of the event based on the time sequence; according to a semantic relationship between the sentence sequence and the plurality of events, context structure information is formed, the context structure information is quantified, a context consistency score is generated, and the context consistency score is used for evaluating logic coherence of the text; the time consistency score and the context consistency score are comprehensively calculated, a time sequence risk index is generated, and the time sequence risk index is used for evaluating the time sequence deviation degree of the text. The problem that the output content of a large model has deviation in multiple dimensions is solved.
Need to check novelty before this filing date? Find Prior Art