Multi-Turn Conversations and Prompts for Rare Disease

Characteristics of This Data Category

Rare disease regulatory data originates from diverse sources. These include policy documents from the National Health Commission and the National Medical Products Administration, clinical practice guidelines, drug inserts, medical insurance catalogs, and various expert consensuses. Documents are typically in PDF, Word, or official web page formats. Update frequency is relatively low, usually quarterly or annually. Document structures are complex, containing extensive professional terminology, legal statutes, medical descriptions, and treatment plans. Fields include drug names, indications, dosage and administration, medical insurance coverage, diagnostic criteria, and genetic characteristics. Units often include milligrams (mg), micrograms (µg), milliliters (mL), international units (IU), and percentages (%), representing medical and pharmaceutical measurement units.

Constraints Imposed by These Characteristics on "Multi-Turn Conversations and Prompts"

The complexity of data sources and update frequency limit the automation level of knowledge base construction, requiring manual intervention for initial screening and annotation. Complex document structures necessitate more refined strategies for segmentation to avoid semantic breaks or information loss. Extensive professional terminology and medical descriptions demand high model comprehension capabilities, potentially leading to misunderstandings or confusion of specialized terms in multi-turn conversations. Fields like medical insurance coverage and diagnostic criteria require high precision; any misinterpretation can affect information accuracy. The mixed use of medical and pharmaceutical units requires prompt design to clearly define unit conversion or recognition strategies, preventing distorted answers due to unit misinterpretation.

Configuration Settings

Configuration Item	Recommended Value	Rationale for Recommendation
`Max Context Window`	`32000 token`	Rare disease regulatory documents are lengthy and information-dense, requiring a larger context window to accommodate more background information.
`Chunk size`	`500-800 characters`	Balances semantic completeness and retrieval efficiency, avoiding noise from overly long segments and information loss from overly short segments.
`Recall count`	`Top 8 entries`	Ensures coverage of different sections from multiple relevant regulatory documents or guidelines, enhancing the comprehensiveness of answers.
`Similarity threshold`	`0.78-0.85`	Professional terminology in the rare disease field has high similarity, requiring a higher threshold to reduce the retrieval of irrelevant content.
`Rerank result count`	`Top 3 entries`	Further refines the most relevant items from the retrieved results, reducing the model's processing burden and improving answer accuracy.
`Maximum Output Length`	`1500 characters`	Regulatory Q&A often requires detailed explanations, ensuring the model has sufficient output space.

Three Common Mistakes

Model output Markdown table formatting is incorrect, and content is truncated. This occurs because the model's output character count exceeds the Maximum Output Length limit, or the Markdown renderer cannot correctly process complex table structures.
In multi-turn conversations, the model cannot accurately distinguish similar symptoms or treatment plans for different rare diseases. This occurs because prompts fail to effectively guide the model to focus on disease-specific diagnostic criteria or differential points.
When users ask about medical insurance coverage, the model fails to return specific payment ratios or reimbursement conditions. This occurs because relevant regulatory documents in the knowledge base are not precisely retrieved, or the prompt does not explicitly request the model to extract such numerical information.

How to Confirm Proper Configuration

Conduct multi-turn conversation tests for typical rare disease regulation queries to check if the model maintains conversational context consistency.
Randomly select complex questions involving medical terminology, units of measurement, and legal provisions to verify the accuracy and professionalism of the model's answers.
Simulate user questions about medical insurance reimbursement, drug dosage, and administration, and cross-reference the model's output with original documents.
Continuously monitor user feedback and periodically adjust Similarity threshold and Rerank result count to optimize retrieval and ranking effectiveness.

The values provided are common starting points and should be measured against the reader's 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.