A multi-modal garment recommendation method

By constructing a knowledge graph and a multimodal knowledge base, combined with a large language model and retrieval enhancement generation technology, the problems of insufficient personalization and monotonous interaction methods in existing clothing recommendation systems are solved, realizing efficient and personalized clothing recommendations that can adapt to rapidly changing fashion trends.

CN120744187BActive Publication Date: 2026-05-29SHENYANG UNIVERSITY OF TECHNOLOGY
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SHENYANG UNIVERSITY OF TECHNOLOGY
Filing Date
2025-06-24
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing clothing recommendation systems rely on users' purchase history or simple preferences, resulting in repetitive and unpersonalized recommendations. They also suffer from limited user interaction methods, lack of real-time feedback and multimodal input support, and difficulty in adapting to rapidly changing fashion trends.

Method used

A knowledge graph is constructed, and combined with a large language model and a multimodal knowledge base, high-dimensional semantic embedding vectors are generated through named entity recognition, entity disambiguation, attribute extraction, and relation extraction. Retrieval enhancement generation technology is used for clothing recommendation, and recommendation scores are calculated by combining the edge weights of the knowledge graph. Finally, structured text and visual content are generated through a large language model.

Benefits of technology

It significantly improves the accuracy and personalization of clothing recommendations, enhances the user interaction experience, can flexibly respond to rapidly changing fashion trends, and supports multimodal data input and real-time feedback.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120744187B_ABST
    Figure CN120744187B_ABST
Patent Text Reader

Abstract

The application discloses a kind of multimodal clothing recommendation methods, first relevant data is constructed as node and edge knowledge graph and stored in Neo4j graph database;Unified multimodal vector representation is extracted to image and text pair, and entity linking is carried out, establish multimodal knowledge base, find the most similar Top-m result with question embedding based on the RAG strategy of knowledge base, and locate the node of the result in knowledge graph, generate subgraph, according to the edge weight in knowledge graph, the recommendation score of each scheme is calculated in combination with subgraph, the final Top-n recommendation result is obtained, the knowledge of Top-n scheme is spliced after user question to generate structured text answer, meanwhile, the text description part is rendered as visual content, to realize the mixed output of combination of text and picture, by the combination of RAG driven by knowledge graph and knowledge base and the generation of high-quality generation model, the application can flexibly capture the semantic information and visual consistency of complex clothing recommendation, thereby significantly improve the accuracy of clothing recommendation.
Need to check novelty before this filing date? Find Prior Art