检索模型训练方法、知识问答方法、装置、设备及介质
By combining a text embedding model and a large model computation module with a preset loss function to optimize the retrieval model, the problems of text relevance and downstream task adaptability in existing knowledge question answering systems are solved, thereby improving the accuracy of retrieval and question answering.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI IFLY DIGITAL TECH CO LTD
- Filing Date
- 2023-09-21
- Publication Date
- 2026-07-17
AI Technical Summary
In existing knowledge-based question-answering systems, retrieval methods primarily focus on text relevance while neglecting the adaptability to downstream tasks, resulting in inaccurate answers and the inability to retrieve correct answers when user question descriptions are vague.
By using a text embedding model and a large model computation module, the similarity and answerability probability values of the training data are obtained. The parameters of the retrieval model are adjusted using a preset loss function threshold. The question elements are analyzed in conjunction with the large model question parsing module to optimize the retrieval model and improve accuracy.
It improves the accuracy of the retrieval model and the accuracy of the question-answering system, solves the problems of text relevance and downstream task adaptability, and reduces the interference caused by vague user question descriptions.
Smart Images
Figure CN117290782B_ABST