A Bridge Specification Injection Method and System Based on Dynamic RAG and Bad Case Feedback
CN123115306APending Publication Date: 2026-11-17CHINA RAILWAY SOUTHWEST SCI RES INST CO LTD
View PDF 0 Cites 0 Cited by
Patent Information
- Application Number
- CN202611069094.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-17
- Publication Date
- 2026-11-17
Smart Images

Figure CN123115306A_ABST
Abstract
This invention discloses a bridge-based specification injection method and system based on dynamic RAG and bad case feedback, belonging to the field of artificial intelligence and natural language processing technology. The method includes automatically extracting multi-dimensional domain labels from reports; before performing vector retrieval, hard filtering of candidate rules in the expert patch library based on domain labels; then calculating cosine similarity of the hard-filtered candidate rules to extract the Top-K most relevant rules and inject them into the large model context; simultaneously, through the bad case feedback mechanism, engineers can extract correct judgment logic into expert meta-rules and make them effective immediately, without modifying code, retraining the model, or restarting the service. This invention fundamentally solves the problem of misattribution across specification systems, achieves low-cost continuous accumulation of expert experience, significantly reduces token consumption, and improves the accuracy and reliability of large-scale model reasoning in vertical domains.
Need to check novelty before this filing date? Find Prior Art
