What data for this category looks like
Data for this category originates from health supervision information systems, medical institution electronic registration systems, complaint and reporting platforms, and law enforcement inspection records, with a monthly update cycle. Daily documents include supervision inspection forms, medical institution practice licenses, rectification notices, supervision opinions, penalty case files, and more. Most documents combine structured fields and unstructured text. Fields cover institution name, license scope, inspection date, rectification requirements, and other details. No unified fixed format exists, and some documents include handwritten annotations or scanned content.
What constraints these characteristics impose on knowledge base retrieval and recall
Monthly updated data requires regular synchronization to the knowledge base to prevent expired information such as institution licenses and rectification requirements from being recalled, which would affect law enforcement accuracy. Mixed retrieval of multiple document types requires distinguishing retrieval logic for structured fields and unstructured text to avoid irrelevant content interfering with core judgments. Law enforcement inspection records and penalty case files contain sensitive compliance information, so recall scope must be strictly controlled to prevent sensitive content leaks. Field differences across different documents require adapting multi-dimensional matching rules during retrieval, ensuring core supervision information such as institution qualifications and personnel qualifications is accurately located, supporting daily time-consuming verification and rectification tracking processes.
How to configure settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches core keywords such as institution qualifications and rectification requirements, filters low-relevance results, and balances retrieval precision and coverage |
limit | Top 10 results | Single retrieval results adapt to the rapid review needs of law enforcement personnel, avoiding excessive information interfering with judgments |
searchMode | Keyword + semantic hybrid | Adapts to mixed retrieval of structured fields and unstructured text, covering exact matching scenarios such as license numbers and inspection dates, as well as semantic matching scenarios such as rectification descriptions |
usingReRank | Enabled | Performs secondary sorting on retrieved results, optimizing relevance ranking for long texts such as penalty case files and supervision opinions |
datasetSearchUsingExtensionQuery | Enabled | Automatically generates extended query terms based on user questions, supplements industry-specific keywords such as "practice license" and "hospital infection inspection", improving retrieval precision |
collectionIds | Specify the exclusive dataset ID for health supervision | Isolates data from other categories, avoids retrieving documents from non-target scenarios, and ensures the retrieval scope only covers data from this category |
Three common configuration errors
- Phenomenon: After upgrading FastGPT to a new version, the knowledge base page displays an abnormal version number or fails to load normally. Cause: The knowledge base configuration file was not synchronized during the upgrade, or the version upgrade script was not fully executed.
- Phenomenon: The number of results returned during knowledge base retrieval does not match the configured
limitparameter. Cause:searchModewas not correctly set to hybrid retrieval mode, resulting in additional filtering of retrieval results for some structured fields. - Phenomenon: Retrieval response time is too long, exceeding a reasonable waiting range. Cause:
collectionIdswas not limited to the exclusive health supervision dataset, causing the retrieval scope to cover a large number of non-target scenario documents, increasing model processing and retrieval time.
How to confirm configurations are correctly applied
- Manually enter a known medical institution name, verify whether the retrieval results include core information of the corresponding document.
- Check the FastGPT retrieval log to confirm that configuration parameters such as
similarityandlimitare correctly called. - Switch to different business questions, such as "Infection control inspection requirements for a certain medical institution", verify whether the relevance of the retrieved results meets expectations.
- Test functions after version upgrade, confirm that the knowledge base page displays normally with no error prompts.
The values provided here are common starting points and should be evaluated 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 2026-09-28.