Multi-turn Conversation and Prompts for Infectious Disease

Data Characteristics

Infectious disease protocol documents typically originate from national health commissions, disease control and prevention centers (CDCs), hospital infection control departments, and medical societies. These documents update frequently, often quarterly or semi-annually, especially with new infectious diseases or drug-resistant strains. Document structures usually include policy regulations, disease definitions, diagnostic criteria, treatment plans, prevention and control measures, and reporting processes. They often include charts, appendices, and references. Fields cover pathogen names, host types, transmission routes, incubation periods, clinical manifestations, laboratory test indicators (e.g., Ct values, antibody titers, gene sequencing results), drug dosages, isolation levels (e.g., N95 mask usage guidelines), and environmental disinfectant concentrations. Units include international units (IU), milligrams (mg), milliliters (ml), time units (hours, days), and percentages (%). Value precision is strictly required.

Constraints on Multi-turn Conversation and Prompts

The high update frequency of infectious disease protocol documents requires an efficient knowledge base synchronization mechanism. This ensures multi-turn conversations use the latest information. Specialized terminology and precise numerical values in documents demand high accuracy from prompts, preventing the model from hallucinating or misinterpreting critical information. Multi-turn conversations must handle complex logical relationships, such as conditional judgments in diagnostic pathways or drug selection and dosage adjustments in treatment plans. These rely on a deep understanding of structured document content. Additionally, similar symptoms and common prevention principles exist across different diseases. Prompts must guide the model to distinguish specific disease details and avoid generalized answers. Precise field and unit information requires the model to accurately reproduce them in responses, especially for drug use or test result interpretation, where any deviation can have serious consequences.

Configuration Settings

Configuration Item	Recommended Value	Rationale
`maxContext`	`8` turns	Infectious disease diagnosis and treatment often involve multi-step reasoning. 8 conversation turns cover most scenarios while controlling computational resource consumption.
`Chunk size` (Segment Length)	`500` characters	Protocol documents often contain detailed technical specifics and operational steps. A segment length of 500 characters helps maintain semantic integrity and avoids truncating critical information.
`Recall count` (Recall Count)	`Top 5` entries	Infectious disease protocols are typically well-structured. The top 5 recall results usually cover key information points for user queries, reducing irrelevant information interference.
`Similarity threshold` (Similarity Threshold)	`0.75`	The domain is highly specialized. A higher matching degree is needed to ensure the accuracy of recalled content and avoid misleading information.
`Rerank result count` (Reranked Return Count)	`3` entries	Building on high-similarity recall, selecting the 3 most relevant document snippets for reranking further improves answer precision.
`temperature`	`0.1`	Protocol Q&A prioritizes accuracy and factuality. A low `temperature` value effectively reduces the risk of model hallucination.

Common Mistakes

In multi-turn conversations, the model repeatedly mentions outdated prevention measures or drug information. This happens when the knowledge base content fails to synchronize with the latest guidelines.
When users ask about specific pathogen detection methods and Ct value ranges, the model returns empty or vague results. This occurs when critical numerical values are separated from their descriptions during original document segmentation, leading to incomplete information in recalled snippets.
When using API calls for Q&A, results differ significantly from online conversations. This happens when stream is set to false, and the detail field is missing because the API call parameter detail was not correctly set to true.

How to Verify Configuration

Select recently updated infectious disease protocol documents. Ask questions about key numbers, units, or diagnostic procedures. Verify the consistency between the model's answers and the document content, especially the precise reproduction of units like mg/kg and IU/ml.
Simulate a doctor's multi-turn follow-up questions during diagnosis, from symptom description to suspected pathogen, then to specific testing plans. Verify the coherence and logical flow of the conversation and check if the maxContext parameter is sufficient.
Ask about differential diagnostic points for various infectious diseases (e.g., bacterial pneumonia versus viral influenza). Observe if the model accurately distinguishes and cites relevant protocol clauses. Evaluate the effectiveness of Similarity threshold (Similarity Threshold) and Rerank result count (Reranked Return Count).

The values provided are common starting points. Measure them against specific 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.