Multi-turn Conversation and Prompts for Rational Drug Use

Data Characteristics in This Domain

Rational drug use regulation data primarily originates from official documents published by national and local health commissions. These include laws, departmental rules, technical specifications, and guidelines. Documents update quarterly or annually, or more frequently if policies change. Formats are often PDF or Word, with some structured text (e.g., HTML pages). Content structure is rigorous, typically including chapters, articles, and appendices. Core fields cover drug names, indications, contraindications, dosage and administration, adverse reactions, interactions, precautions for special populations, pharmacist review processes, and prescription review standards. Data frequently contains medical terminology, generic and brand drug names, dosage units (e.g., mg/kg, IU), and time units (e.g., hours, days).

Constraints Imposed by These Characteristics on "Multi-turn Conversation and Prompts"

The rigorous and specialized nature of rational drug use regulation documents requires multi-turn dialogue systems to accurately identify medical terminology and drug information when understanding user intent. Document update frequency dictates the timeliness of knowledge base retrieval; outdated regulations can lead to incorrect medication advice. Complex document structures, such as nested clauses and appendices, demand higher requirements for text segmentation and indexing strategies to ensure complete retrieval of relevant information. Dosage and time units in fields necessitate prompt design that guides the model to perform correct numerical inference and comparison, preventing safety issues due to unit confusion. In multi-turn conversations, users may progressively add information, requiring the system to accumulate context and make judgments based on the latest complete information.

Configuration Settings

Configuration Item	Suggested Value	Rationale
`maxContext`	800–1200 characters	Ensures critical information from multi-turn conversations is accommodated while controlling model inference costs
`Recall Count`	5–8 items	Covers multiple related clauses in regulations, improving information comprehensiveness
`Similarity Threshold`	0.75–0.85	Balances recall precision and recall rate, avoiding interference from irrelevant clauses
`Rerank Return Count`	Top 3 items	Prioritizes the most relevant and important regulatory clauses, enhancing user efficiency in obtaining key information
`Segment Length`	300–500 characters	Adapts to longer clause descriptions in regulatory documents, maintaining semantic integrity
`temperature`	0.3–0.5	Reduces the randomness of model-generated content, ensuring the rigor and accuracy of responses

Three Common Mistakes

Responses lack timeliness, presenting abolished or updated regulatory clauses. This occurs because the knowledge base is not updated promptly, or index rebuilding did not cover the latest documents.
During a conversation, the model fails to correctly understand unit information in drug dosages or administration, providing vague or incorrect advice. This happens because prompts do not explicitly emphasize unit recognition, or the knowledge base does not standardize units.
In multi-turn conversations, the model fails to effectively utilize background information provided by the user in previous turns, leading to repetitive questions or contradictory responses. This is due to improper context management configuration, where the maxContext parameter is insufficient to carry the complete conversation history.

How to Verify Configuration

Select recently updated rational drug use regulation documents. Ask multi-turn questions about their core clauses, checking if responses are entirely based on the latest version.
Design complex questions involving specific dosages and administration frequencies. Verify if the system can accurately cite and process these values and units in its responses.
Simulate scenarios where users progressively provide symptoms, medical history, and medication contraindications. Observe if the system can provide more precise medication advice after each follow-up question, combining all context, and check maxContext usage in logs.

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.