Data Characteristics
Data for attenuated inactivated vaccine regulations and SOP documents primarily originates from regulatory files published by the National Medical Products Administration (NMPA) and the Pharmacopoeia Commission, guidelines from industry associations, and internal quality management system documents. These documents are typically in PDF, Word, or scanned image formats. Update frequency is relatively low, generally quarterly or annually, with additional updates for major policy changes or technical breakthroughs. Document structure is chapter-based, including fixed modules such as introduction, definitions, responsibilities, operating procedures, quality control, and record requirements. Fields involve vaccine batch numbers, production dates, expiration dates, storage conditions, and test results. Units strictly follow national standards, such as IU/mL, ℃, and h.
Constraints on Deployment and Upgrade
The data characteristics of attenuated inactivated vaccine regulation documents impose specific constraints on FastGPT's deployment and upgrade. First, the semi-structured nature of documents and the presence of scanned images require stable and reliable OCR and document parsing capabilities for accurate text extraction. The low update frequency means significant document import during initial deployment, but subsequent incremental update pressure is low, focusing mainly on identifying revisions and version management. Second, strict field and unit requirements necessitate attention to entity recognition and numerical extraction precision during knowledge base construction to support accurate question answering. During system upgrades, ensure compatibility between the existing knowledge base's semantic indexing and vector embedding models to prevent recall accuracy degradation due to model updates. Furthermore, compliance requirements mandate strict validation for every knowledge base update and model adjustment to guarantee the authority and accuracy of question answering results.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
UPLOAD_FILE_MAX_SIZE | 200 MB | Individual regulation files or SOP documents can be large, especially PDFs with charts or scanned pages. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | OCR parsing for large PDFs or scanned images takes a long time. Allocate sufficient time to prevent parsing failures. |
Chunk size | 800–1200 characters | Regulatory and SOP documents have strong logical coherence. Longer segments help retain contextual information and improve recall completeness. |
Recall count | Top 5 entries | Accuracy is critical for regulatory questions. Increasing the number of recalled items appropriately covers more relevant content and improves hit rates. |
Similarity threshold | 0.85 | Vaccine regulation questions demand high precision. A high similarity threshold effectively filters irrelevant information and ensures the authority of returned content. |
Rerank result count | Top 3 entries | After re-ranking, the top 3 most relevant items are sufficient for most precise question-answering needs, while avoiding information overload. |
Common Pitfalls
- Some document content is missing or garbled after knowledge base import. This is due to insufficient text recognition capability of the document parser for complex tables or images, or incorrect OCR language pack configuration.
- The system returns "no relevant content found" after a user query. This might be because of an unreasonable knowledge base segmentation strategy, leading to key information being split, or the vector model not fully capturing the semantics of specialized terms.
- Question answering accuracy decreases after upgrading FastGPT. This is due to adjustments in the new version's embedding model or retrieval algorithm, causing compatibility issues with the vector representations of the old knowledge base.
Verification of Configuration
- Select 10 typical attenuated inactivated vaccine regulation documents. Upload them to the knowledge base. Check if all document parsing statuses are "success." Randomly sample document previews to verify consistency with the original text.
- Test at least 20 questions involving key fields such as vaccine batch numbers, storage temperatures, and expiration dates. Verify if the system can accurately extract corresponding values and units, and cross-reference with the original text.
- Simulate a user query such as "What are the storage conditions for attenuated inactivated vaccines?" Confirm if the system's response includes key information like
2–8℃and compare it with the original regulation text to verify information accuracy.
Note: The values provided are common starting points. Measure them against your own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.