A method for cross-scenario data migration and adaptation of a life service intelligent agent

By constructing a common space and a cross-domain attention network and dynamically adjusting weights, the cold start problem and feature semantic drift caused by data fragmentation across scenarios are solved, thereby improving recommendation accuracy and user experience.

CN122089438APending Publication Date: 2026-05-26GUANGDONG GUOLI EDUCATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
GUANGDONG GUOLI EDUCATION TECH CO LTD
Filing Date
2026-02-13
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Data fragmentation across scenarios leads to difficulties in cold starts in new scenarios, feature semantic drift, and difficulty in adapting static rules, affecting recommendation accuracy and user experience.

Method used

By constructing a common space, standardizing source data, weighted migration, closed-loop optimization, and conflict circuit breaking, efficient migration and adaptation of user preference data from the source domain to the target domain is achieved. Cross-domain attention networks are used to dynamically adjust weights, and the model is optimized by combining gradient descent and binary cross-entropy loss function.

Benefits of technology

It significantly improves the accuracy of recommendations in new scenarios, solves the problem of feature semantic drift, and enhances the user experience and the flexibility and adaptability of the recommendation system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122089438A_ABST
    Figure CN122089438A_ABST
Patent Text Reader

Abstract

This invention provides a cross-scenario data migration and adaptation method for intelligent agents in life services. The method includes extracting common dimensions across different scenarios to construct a shared feature vector space; converting unstructured labels from the source scenario into mathematical vectors; calculating the influence of source domain data on the target scenario and constructing a cross-domain attention network; based on an initial migration strategy obtained through weighted migration, adjusting the weights in reverse according to real user feedback data in the target scenario; defining a loss function and a comprehensive evaluation index, and updating parameters using gradient descent; when new and old data conflict, setting a time decay factor based on the principle of prioritizing real-time data, and if behavioral records exist in the target scenario within a specified time period, forcibly setting source domain weights and directly adopting the target domain profile. This invention achieves efficient migration and adaptation of user preference data from the source domain to the target domain through multiple stages, including constructing a common space, standardizing source data, weighted migration, closed-loop optimization, and conflict circuit breaking.
Need to check novelty before this filing date? Find Prior Art