一种用于推荐系统的敏感属性遗忘方法与装置
By decomposing the user embedding matrix in the recommender system and optimizing irrelevant vectors using the HSIC kernel, the high overhead and performance loss problems in the prior art are solved, achieving low overhead, high compatibility and flexible sensitive attribute forgetting, while maintaining the performance and privacy protection of the recommender system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HANGZHOU DIANZI UNIV
- Filing Date
- 2026-04-10
- Publication Date
- 2026-07-17
AI Technical Summary
Existing recommendation systems suffer from high computational overhead and performance loss when it comes to forgetting sensitive attributes, making it difficult to flexibly handle diverse forgetting requirements in deployed models.
By performing matrix decomposition on the user embedding matrix, orthogonal bases of relevant and irrelevant spaces are obtained, decoupling the user representation vector into relevant and irrelevant vectors. The Hilbert-Schmidt independence criterion (HSIC kernel) is then used to construct a forgetting loss function to optimize the irrelevant vectors to update the user representation vectors while keeping the relevant vectors unchanged.
It achieves low-overhead, high-compatibility, and flexible forgetting of sensitive attributes, maintains the performance stability of the recommendation system, and adapts to diverse forgetting needs.
Smart Images
Figure CN121996847B_ABST