Vector retrieval method, device, storage medium and program product for multi-service scenario

By receiving search requests and generating search vectors through data center services, the search scope and similarity of the material library are determined, which solves the problems of storage resource waste and computing overhead in multiple business scenarios, realizes cross-scenario joint search, and improves search efficiency and flexibility.

CN121301445BActive Publication Date: 2026-06-05ZHEJIANG BAORONG MEDIA TECH (ZHEJIANG) CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Patents(China)
Current Assignee / Owner
ZHEJIANG BAORONG MEDIA TECH (ZHEJIANG) CO LTD
Filing Date
2025-12-11
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

In existing technologies, storage resources are wasted and vectorized computing overhead increases in multiple independent business scenarios, making it impossible to achieve cross-scenario joint retrieval and limiting the breadth and flexibility of retrieval.

Method used

The system receives target retrieval requests through data center services, determines modality types and generates retrieval vectors, defines the retrieval scope in a pre-defined material library, identifies candidate material identifiers based on similarity calculation results, and uses their content as retrieval results. It also optimizes storage and computation using hierarchical storage and lazy loading mechanisms.

Benefits of technology

It reduces storage resource waste and vectorized computation overhead, enables cross-scenario joint retrieval, and improves the breadth and flexibility of retrieval.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121301445B_ABST
    Figure CN121301445B_ABST
Patent Text Reader

Abstract

The application discloses a kind of multi-service scene vector retrieval method, equipment, storage medium and program product, it is related to big data technical field.The method comprises: receiving target retrieval request, determining the modal type of target retrieval request, and determining the target retrieval vector matched with target retrieval request based on modal type;In the target material library determined in advance, the search range is determined, and candidate material identification set is determined;Wherein, target material library includes the material of multiple service scenes and the feature vector corresponding to each material;The similarity of target retrieval vector and the feature vector of each candidate material in candidate material identification set is determined respectively;Based on similarity calculation result, target candidate material identification is determined, and the material content corresponding to target candidate material identification is determined as the search result of target retrieval request.The scheme of the application can reduce storage resource waste and vectorization calculation overhead, realize cross-scene joint retrieval, and improve the breadth and flexibility of retrieval.
Need to check novelty before this filing date? Find Prior Art