一种地址模型的训练方法、装置、设备及存储介质
By acquiring a training address encoding dataset and adjusting the network parameters of the address model, the problem that existing language models are not suitable for processing address languages is solved, and the address model is able to learn address-related feature information, supporting downstream address services.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING JINGDONG 360 DEGREE E COMMERCE CO LTD
- Filing Date
- 2022-05-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing text language processing models are not suitable for processing address languages because the relationships in address languages are mainly reflected in regional affiliation and spatial distance, rather than in the contextual content of general natural languages.
By acquiring a training address encoding dataset, the training address encoding data is input into the initial address model for each preset address service, and a predicted address vector representation is output. The network parameters of the initial address model are then adjusted based on the predicted address vector representation and standard address service data, thus solving the problem that existing language models are not suitable for processing address languages.
This enables the initial address model to learn address-related feature information, and the output predicted address vector representation can support downstream address services, thus broadening the applicable scenarios of the address model.
Smart Images
Figure CN117195993B_ABST