An unstructured single document full-chain intelligent processing method and system based on multi-modal constraint and completeness score and a medium

By using intelligent document classification based on semantic fingerprint features and keyword topology mapping, combined with a large language model and sparse data completion algorithm, the problems of poor adaptability and noise interference in unstructured document processing are solved, achieving efficient and accurate document processing and data output.

CN121705428BActive Publication Date: 2026-06-12KUNSHAN EAST CHINA INFORMATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Patents(China)
Current Assignee / Owner
KUNSHAN EAST CHINA INFORMATION TECH CO LTD
Filing Date
2026-02-13
Publication Date
2026-06-12

AI Technical Summary

Technical Problem

Existing technologies face problems such as poor adaptability, weak resistance to noise interference, and inability to handle cross-page extraction of long documents when processing unstructured documents, resulting in high operation and maintenance costs and low processing efficiency.

Method used

By extracting the semantic fingerprint features of file names and combining them with a keyword topology mapping table to determine the document type, decoupling document groups and constructing a task-specific processing link, extracting content page by page using a large language model, and combining a sparse data differential completion mechanism to achieve structured correction and logical verification.

Benefits of technology

It achieves high-precision document type determination and intelligent aggregation, improves the accuracy and efficiency of data processing, reduces operation and maintenance costs, and solves the fault tolerance problem under cross-page extraction and multi-source heterogeneous data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121705428B_ABST
    Figure CN121705428B_ABST
Patent Text Reader

Abstract

The application relates to a non-structured single document full-chain intelligent processing method and system based on multi-modal constraint and completeness scoring and a medium, and relates to the technical field of intelligent document processing. The non-structured single document full-chain intelligent processing method comprises the following steps: extracting semantic fingerprint features of a file name, combining a preset keyword topology mapping table, and determining a document type; according to the document type, combining the file name in the current document, aggregating all files, and constructing a plurality of document groups; decoupling the document groups, constructing corresponding task-specific processing links, and extracting file content and file formats of each file page by page; according to a commodity serial number, combining a structured mode constraint layer, correcting the file format, constructing a field alias mapping library, combining field weight, and calculating commodity record completeness; according to the commodity record completeness, combining a sparse data difference completion mechanism, structurally correcting and logically verifying the file content; and by decoupling business knowledge into configurable knowledge nodes, the operation and maintenance cost is greatly reduced.
Need to check novelty before this filing date? Find Prior Art