Enterprise low-carbon evaluation method and system based on search enhancement generation

By using physical coordinate-based table semantic awareness and counterfactual consistency verification, the problem of accurately extracting cross-page table data and adaptive unit alignment in enterprise low-carbon reports has been solved, achieving high-precision and traceable low-carbon assessment.

CN121836134BActive Publication Date: 2026-06-16ZHEJIANG UNIV CITY COLLEGE
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
ZHEJIANG UNIV CITY COLLEGE
Filing Date
2026-03-16
Publication Date
2026-06-16

AI Technical Summary

Technical Problem

Existing technologies struggle to accurately extract carbon emission data from corporate low-carbon reports, especially since the semantic integrity of cross-page tables is compromised, and there is a lack of adaptive unit alignment and counterfactual consistency checks, leading to inaccurate and unreliable evaluation results.

Method used

Table data is reconstructed using physical coordinate-based semantic perception technology, combined with a dynamic carbon emission factor library for unit conversion, and the accuracy of the results is ensured through a counterfactual consistency verification mechanism, including a hybrid retrieval strategy and traceable reasoning from a large language model.

Benefits of technology

It achieves high-precision and traceable low-carbon assessment, ensuring high recall and accuracy of carbon emission data, and meeting the stringent requirements of financial institutions and auditing firms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121836134B_ABST
    Figure CN121836134B_ABST
Patent Text Reader

Abstract

The application discloses a kind of based on retrieval enhancement generation's enterprise low carbon evaluation method and system, method includes: by pre-processing module, non-structured report is carried out format identification and table semantic reconstruction, in combination with carbon accounting standard, data cleaning and unit standardization are carried out, and vector index library is constructed;In the retrieval stage, evaluation index is converted into query vector, and relevant evidence is accurately positioned by hybrid retrieval;Enhanced prompt word is guided to build the logic of unit standardization and infer large model, and structured intermediate result is generated;Introduce counterfactual consistency checking mechanism, compare generated content with original context, and eliminate low confidence result;Finally, combined with scoring rules and evidence confidence, weighted calculation is carried out, and enterprise low carbon level score is obtained.The application realizes high-precision, traceable automatic low-carbon evaluation by fusing table semantic perception based on physical coordinates, embedded industry standard unit normalization reasoning and counterfactual consistency checking mechanism.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Large model illusion correction method and device based on iterative evaluation algorithm

    CN118709777A

  • Method, system and equipment for large language model fact verification based on retrieval enhancement and medium

    CN119204025A