Deployment and Upgrade for Respiratory System Protocols

Data Characteristics

Medical institutions or professional associations typically publish respiratory system diagnosis and treatment protocols and SOP documents. Updates are infrequent, usually semiannually or annually, with occasional temporary updates for major epidemics or treatment breakthroughs. Documents are often in PDF, Word, or internal hospital knowledge base pages. Their structure is rigorous, containing extensive medical terminology, diagnostic criteria, treatment procedures, drug dosages, operational specifications, and ethical requirements. Diagnostic criteria sections often include disease classification codes (e.g., ICD-10 codes) and normal/abnormal thresholds for laboratory indicators. Drug dosages are precise, down to milligrams (mg) or micrograms (μg), specifying administration routes. Operational specifications typically include detailed step lists, risk warnings, and emergency procedures.

Constraints on Deployment and Upgrade

Infrequent updates of respiratory system protocol documents allow for longer knowledge base reconstruction or incremental update cycles, reducing resource consumption from frequent data synchronization. The rigorous document structure and extensive medical terminology require maintaining medical concept integrity during segmentation to prevent semantic deviations from misinterpretation. Precise dosages and indicator thresholds demand high accuracy in text extraction and entity recognition to avoid numerical errors. Diverse document formats (PDF, Word) necessitate robust file parsing capabilities. Standardized fields like ICD-10 codes or generic drug names are crucial for precise recall and cross-validation, requiring preprocessing or special tagging before vectorization.

Configuration Settings

Configuration Item	Recommended Value	Rationale
`UPLOAD_FILE_MAX_SIZE`	`500 MB`	Respiratory system protocol documents are often long; a single file may contain multiple SOPs, requiring support for larger file uploads.
`Chunk size` (Segment Length)	`800–1200 characters` (characters)	Dense medical terminology requires longer segments to maintain contextual integrity and prevent truncation of key information.
`Recall count` (Recall Count)	`Top 8 entries` (top 8)	Protocol Q&A demands high accuracy; increasing recall count improves the probability of selecting highly relevant segments.
`Similarity threshold` (Similarity Threshold)	`Calibrate by actual measurement` (calibrate by actual measurement)	Adjust based on specific document content and user query habits using a test set to ensure recall precision.
`Rerank result count` (Rerank Return Count)	`Top 5 entries` (top 5)	Rerank based on initial recall to further prioritize the presentation of the most relevant content.
`PARSE_FILE_TIMEOUT_SECONDS`	`600 seconds` (seconds)	Parsing large PDF or Word documents can be time-consuming; allow sufficient time to prevent parsing timeouts.

Common Pitfalls

Knowledge base document parsing fails, displaying file parsing error or file processing timeout. This occurs when complex document formats, including numerous images, tables, or embedded objects, are not correctly handled by the default parser, or PARSE_FILE_TIMEOUT_SECONDS is set too low.
When users query specific disease treatment dosages, the model's answers are inaccurate or lack critical numerical values. This happens when segments are too short, separating dosage information from related context like drug names or administration routes, leading to loss of association during vectorization.
After updating the knowledge base, questions about old protocol versions still appear, or new protocol content is not fully covered. This indicates that the incremental update strategy did not effectively identify and replace all relevant old documents, or indexing was incomplete.

Verification Steps

Upload typical documents (e.g., PDFs containing ICD-10 codes and drug dosage tables). Check the Knowledge Base Management page to confirm the file status is completed and that the segment count is consistent with the document length.
Ask questions about specific operational steps or diagnostic criteria from the document. Verify that the Q&A results accurately cite the original passages and check if the cited Recall count (recall count) meets expectations.
Review Q&A records in Log Management. Analyze the recall similarity distribution to ensure most query similarities are within a reasonable range; excessively low or high similarity may indicate configuration issues.
Simulate the protocol update process. Upload a new version of the document and perform queries. Verify that old content is effectively replaced by the new version and check if key fields like version number or publication date are correct.

Note: The values provided are common starting points. Measure against specific samples to determine optimal settings.

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.