业务文档检索方法、装置、设备及存储介质
By implementing interactive learning and multi-attention mechanism encoding in a dual-tower model architecture, the problems of low efficiency and low accuracy in business document retrieval in the financial field are solved, achieving more efficient and accurate document retrieval.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PING AN TECH (SHENZHEN) CO LTD
- Filing Date
- 2023-06-12
- Publication Date
- 2026-07-17
AI Technical Summary
Existing business document retrieval methods suffer from low retrieval efficiency or low accuracy in the financial field.
A dual-tower model architecture is adopted for interactive learning, and a multi-attention mechanism is combined to perform vector encoding on the set of documents to be retrieved in the financial field. The vector representation capability of the model is improved through multi-stage training, including multi-stage general training of a pre-built information extraction model to obtain a standard dual-tower model, and multi-attention dense vector encoding and query vector encoding are performed using a standard document model, and finally document retrieval is performed.
It improves the accuracy and efficiency of document retrieval, especially in the financial field, where a single document can better match different queries, enhancing retrieval accuracy while maintaining the retrieval efficiency of the traditional dual-tower model.
Smart Images
Figure CN116662488B_ABST