User geographical position positioning method and device, terminal equipment and storage medium
By constructing a multi-dimensional user relationship graph and adopting a multi-scale feature fusion model using graph attention networks, the problem of inaccurate user geolocation in existing technologies is solved, achieving more efficient and accurate user geolocation prediction.
Patent Information
- Application Number
- CN202510540248.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-27
- Publication Date
- 2026-06-26
- Estimated Expiration
- 2045-04-27
AI Technical Summary
Existing technologies fail to fully capture the geographic distribution characteristics of users in user geolocation, resulting in low positioning accuracy and efficiency, especially in social network graphs where high-order neighbors with close geographical proximity but weak social connections are ignored.
By acquiring target social data of social users, a multi-dimensional user relationship graph is constructed using a pre-established geolocation prediction model. A multi-scale feature fusion model based on graph attention networks, including a dual-channel graph encoder and a cross-gated feature fusion algorithm, is then used to process the social network graph and location association graph to obtain the overall representation and location label of the user.
It improves the accuracy and efficiency of user location prediction by constructing social network graphs and location association graphs through social relationships and latent location correlations, thus achieving more comprehensive user geolocation positioning.
Smart Images

Figure CN120429750B_ABST