Smart home reasoning data optimization method and related device
By extracting feature samples from real interaction records and generating targeted synthetic data, and combining bucket sampling and LLM cleaning model to optimize the smart home inference dataset, the problem of low data quality was solved, the diversity and accuracy of the dataset were improved, and the generalization ability of the model was enhanced.
CN122153251APending Publication Date: 2026-06-05MIDEA GRP (SHANGHAI) CO LTD +1
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- MIDEA GRP (SHANGHAI) CO LTD
- Filing Date
- 2026-03-10
- Publication Date
- 2026-06-05
Smart Images
Figure CN122153251A_ABST
Abstract
Embodiments of the present application provide a smart home reasoning data optimization method and related equipment. The smart home reasoning data optimization method comprehensively improves the authenticity, diversity and label accuracy of the smart home reasoning data set by fusing the construction mode of real and synthetic data, the multi-dimensional bucket sampling mechanism and the closed-loop data quality optimization pipeline, thereby significantly enhancing the generalization ability and reasoning reliability of the downstream model in a complex real environment.
Need to check novelty before this filing date? Find Prior Art