A lightweight intelligent labeling method for multi-modal data of small and medium-sized e-commerce commodities

By employing an adaptive parsing and lightweight model-driven hierarchical annotation method, the multimodal data annotation challenge of small and medium-sized e-commerce platforms has been solved, achieving efficient and accurate product data annotation, reducing costs and technical barriers, and improving operational efficiency and user experience.

CN122114843APending Publication Date: 2026-05-29GUANGDONG UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
GUANGDONG UNIV OF SCI & TECH
Filing Date
2026-02-03
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Small and medium-sized e-commerce platforms face problems such as high cost, low efficiency, high technical threshold, poor data adaptability and serious redundancy deviation in the labeling of multimodal data of products. Existing technologies are difficult to meet their needs for rapid digital operation.

Method used

We employ a multimodal data adaptive parsing, lightweight model-driven approach, and hierarchical annotation collaboration method to construct an adaptive parsing module, a lightweight model, an e-commerce-specific dictionary, and a hierarchical annotation system. Combined with artificial intelligence algorithms for data processing, we achieve efficient and accurate multimodal data annotation for products.

Benefits of technology

It reduces labeling costs and time investment, simplifies operation processes, adapts to the resource limitations of small and medium-sized e-commerce businesses, improves labeling efficiency and accuracy, supports intelligent product classification and recommendation, and enhances user experience and business conversion results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122114843A_ABST
    Figure CN122114843A_ABST
Patent Text Reader

Abstract

The application discloses a kind of light-weight intelligent labeling methods for commodity multi-modal data of small and medium e-commerce, which comprises the following steps: self-adaptive analysis, format normalization and redundancy filtering of commodity original multi-modal data;Special labeling dictionary for e-commerce is constructed, and image attribute recognition and text key information extraction model are trained based on light-weight model respectively;Through the layered labeling mode of "core layer full-automatic labeling" combined with "extension layer semi-automatic labeling", "model pre-labeling + artificial light-weight review" is used for collaborative processing, and adaptive matching of labeling template is realized;The labeling result is automatically checked in multiple dimensions, and the model and dictionary are continuously optimized through incremental learning;Support multi-format labeling result output and visual preview.The application reduces the labeling cost and technical threshold of small and medium e-commerce significantly while ensuring the accuracy of labeling through light-weight model and layered collaborative strategy, and improves the efficiency of commodity data management.
Need to check novelty before this filing date? Find Prior Art