What this category of data looks like
Data for this category primarily comes from health supervision information systems, medical institution electronic registration systems, complaint reporting platforms, and law enforcement inspection records. Daily documents include supervision inspection forms, medical institution practice licenses, rectification notices, supervision opinions, penalty case files, and others. These cover two structural types: structured forms and official documents. Fields include institution name, license scope, rectification items, penalty basis, and more. No unified standardized unit field is configured. The data update frequency is monthly.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The monthly update frequency requires knowledge base synchronization tasks to run on a natural month cycle, to ensure retrieved data matches current regulatory status. The mixed structure of multiple document types requires support for precise matching of structured form fields and full-text recall of unstructured official documents. The lack of a unified standardized unit field requires avoiding invalid matching related to units. The unstructured nature of law enforcement records and complaint data requires retrieval support for long text segment recall. In the party building learning Q&A scenario, both compliance basis and learning reference content must be recalled, so association recall rules for multi-source data must be configured.
How to configure the settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets strict retrieval requirements for matching compliance-related documents, filters low-relevance non-regulatory results |
limit | Top 8 entries | Covers multiple types of information such as institution qualifications, rectification status, and law enforcement records that need verification during a single supervision inspection |
searchMode | keyword + semantic | Balances precise keyword matching for structured fields such as license numbers and rectification items, and semantic matching for unstructured content such as law enforcement scenario descriptions |
usingReRank | Enabled | Performs secondary sorting of recalled mixed-format documents by relevance, optimizes sorting effects for long texts and structured forms |
datasetSearchUsingExtensionQuery | Enabled | Expands party building-related compliance expressions based on the original question, adapts to scenario association requirements for party building learning Q&A |
collectionIds | Configure exclusive collection IDs grouped by business | Distinguishes health supervision business data from party building learning data, avoids interference from cross-domain retrieval results |
Three common mistakes
- Phenomenon: Retrieval results are empty after uploading supervision inspection forms or rectification notices in HTML format. Cause: The FastGPT built-in HTML text extraction plugin is not enabled, and default parsing rules do not support HTML format documents.
- Phenomenon: When calling the knowledge base API to retrieve, only the previously mentioned abbreviated name of the medical institution is used as the query, and the corresponding practice license document cannot be recalled. Cause: Question expansion logic based on historical conversations is not configured, so reference resolution cannot be completed.
- Phenomenon: After triggering knowledge base retrieval, system GPU resources are not called, and retrieval time exceeds expectations. Cause: GPU computing resources are not bound to the retrieval node, or GPU-accelerated embedding model calls are not enabled.
How to confirm the configuration is complete
- Upload 1 HTML document of a supervision inspection form, perform a retrieval test, and confirm that the corresponding parsed text result can be returned.
- Call the API and pass a historically associated question containing the abbreviated institution name, confirm that the corresponding practice license document can be recalled.
- Check the resource usage of the retrieval node, confirm that GPU resources are being called normally.
- Adjust the relevant parameter ranges, verify that the relevance matching degree of retrieval results meets business requirements.
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 2026-09-28.