Source Citation and Traceability for Infectious Disease

Data Characteristics

Infectious disease protocols and SOP documents originate from national health commission guidelines, CDC prevention and control plans, hospital infection control policies, and professional society consensus statements. These documents are typically PDFs, Word files, or structured XML. Update frequency varies: national guidelines may update annually or every few years, while internal hospital policies revise quarterly or semi-annually based on operational needs and directives. Document content is highly specialized, containing extensive medical terminology, disease classification codes (e.g., ICD-10), drug dosage units (mg/kg, IU), microbial test results (CFU/mL), and complex diagnostic flowcharts and treatment pathways. Cross-references are common, such as a protocol citing specific clauses from a national guideline.

Constraints on Source Citation and Traceability

The specialized nature and frequent revisions of infectious disease protocol documents demand high accuracy and timeliness in source citations. Complex medical terminology and units can cause standard text segmentation methods to split critical information, impacting recall quality. Cross-references mean a single text chunk might not provide complete context, requiring the system to identify and link multiple reference points. Frequent updates challenge knowledge base synchronization and version management, ensuring users receive the latest citation information to avoid decisions based on outdated protocols. Precise traceability to the original page number or paragraph is a fundamental compliance requirement in the medical field. Citation results must point to the document and precisely locate the specific position within it.

Configuration Settings

Configuration Item	Recommended Value	Rationale
`Chunk size` (Chunk Size)	`500–800 characters`	Balances context completeness and retrieval efficiency; avoids splitting key medical concepts.
`Recall count` (Recall Count)	`Top 5–8`	Increases the probability of recalling relevant passages, especially for complex queries.
`Similarity threshold` (Similarity Threshold)	`0.75–0.85`	Ensures professional relevance of recalled content, filtering low-quality matches.
`Rerank result count` (Reranked Return Count)	`Top 3`	Provides concise core citations while maintaining quality, reducing redundancy.
`Citation source display format` (Citation Display Format)	`Filename + Page Number / Paragraph`	Meets medical field compliance requirements for precise traceability.
`Citation Variable Name`	`quote_docs`	Facilitates identification and processing of cited document information in API calls.

Common Mistakes

The quote_docs field is empty after an API call, or the returned citation filename is None. This usually indicates that the citation feature is not correctly enabled in the knowledge base configuration, or metadata extraction failed during document processing.
The content cited in the model's response deviates from the original text or points to irrelevant paragraphs. This may be due to a Similarity Threshold set too low, recalling semantically irrelevant text blocks, or a Chunk Size that is too large, diluting core information.
The citation source displays the correct filename but cannot precisely locate the page number or specific paragraph. This often occurs when original documents were not structurally parsed during knowledge base import, or the parser failed to recognize page/paragraph markers.

Verification Steps

For typical infectious disease protocol questions, check if the quote_docs field in the model's response contains the correct filename and page number/paragraph information.
Randomly select 5–10 question-answer pairs. Verify that the cited text snippets are semantically consistent with the original document content and can be precisely located.
After a document update, test if the knowledge base's citation sources have synchronized to the latest version and correctly cite the updated content.
Use queries containing specific medical terminology or ICD-10 codes. Check if the recalled citation sources accurately cover these keywords.

The values provided are common starting points. Measure them against your own 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.