信息处理方法及装置、电子设备和存储介质
By generating item feature vectors through image encoders and text encoders, and combining them with dynamically updated reference features, the inefficiency caused by manual labeling of item information on e-commerce platforms is solved, enabling fast and efficient identification of item categories and attributes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JINGDONG TECH HLDG CO LTD
- Filing Date
- 2022-08-24
- Publication Date
- 2026-07-17
AI Technical Summary
The labeling of item information on existing e-commerce platforms relies on manual input and review, which consumes a lot of manpower and is prone to errors, and is inefficient in the process of frequent updates to item categories and attributes.
The system generates item feature vectors by using an image encoder and a text encoder, and dynamically updates them by combining them with reference features from the item information database. This automatically determines the item category and attributes, and trains the item feature extraction model and category feature model using a contrastive learning loss function.
It can quickly identify item categories and attributes without the need to build a dataset, improving identification efficiency, and supports dynamic updates of item categories and attributes to adapt to rapidly changing item management scenarios.
Smart Images
Figure CN115374862B_ABST