What the data for this category looks like
Data sources for this category include the national medical insurance information platform, hospital HIS and medical insurance interfaces, medical insurance catalog and coding library, and cross-provincial medical settlement platform. Data updates run daily. Daily processed documents cover medical insurance settlement lists, expense details, medical record homepages, prescriptions, rejection payment lists, and settlement reconciliation statements. Document fields involve diagnosis and treatment project codes, medical consumables codes, expense amounts, insured person identity identifiers, settlement status, and more. Some documents require linking structured codes and unstructured medical record text. Units include RMB yuan, code numbers, person-times, and similar units.
What constraints these characteristics impose on knowledge base retrieval and recall
Because data sources are scattered and updated daily, a scheduled synchronization mechanism must be configured to ensure the knowledge base covers the latest medical insurance catalog and settlement data. Documents contain structured code fields and unstructured medical record text. Retrieval must balance precise matching of diagnosis and treatment project and medical consumables codes, and semantic matching of expense and medical record consistency. For single or batch audit scenarios, support must be provided for single retrieval to link multiple types of documents, while avoiding redundant results that interfere with audit judgments. Additionally, strict matching requirements for professional code fields mean retrieval modes must balance exact matching and semantic recall, to adapt to query needs across different audit stages.
Configuration settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Balances precise matching for code fields and semantic recall, avoids low-match irrelevant results interfering with audits |
limit | Top 10 entries | Controls the number of recall results per retrieval, adapts to quick review needs for single settlement audits. Adjust based on actual testing for batch scenarios |
searchMode | Hybrid retrieval | Balances exact matching for structured codes and semantic recall for unstructured medical records, covers multi-dimensional query needs for settlement audits |
usingReRank | Enabled | Re-ranks recall results, prioritizes documents linked to settlement codes and expense details, improves audit efficiency |
datasetSearchUsingExtensionQuery | Enabled | Automatically optimizes query statements, supplements synonyms or standard expressions related to medical insurance codes, adapts to retrieval needs for professional terminology |
collectionIds | Collection IDs divided by business modules | Accurately links dedicated knowledge base collections for medical insurance catalogs, settlement documents, medical record archives, and similar content, avoids interference from cross-category documents |
Three common configuration mistakes
- Incorrect return format when calling the knowledge base retrieval interface, with a prompt that document types are not supported. Cause: Failed to correctly configure parsing rules for medical insurance settlement docx documents, or failed to link the corresponding business collection when uploading documents.
- Mismatch between the number of retrieval results and expected values, with results exceeding or falling below the set range. Cause: Failed to correctly set the value of the
limitparameter, or failed to filter non-audit related documents in hybrid retrieval mode. - Failure to retrieve matching diagnosis and treatment project codes in retrieval results. Cause:
datasetSearchUsingExtensionQueryis not enabled, or thesimilaritythreshold is set too high, causing exact matching code documents to not be recalled.
How to confirm correct configuration
- Upload a medical insurance settlement list docx document to the corresponding business collection, call the retrieval interface to query the codes within this document, verify that the returned results include this document.
- Adjust the
similarityparameter to 0.8, retrieve high-frequency audit query terms, check whether the matching degree of returned results meets expectations. - After enabling
usingReRank, compare the result order before and after re-ranking, confirm that highly relevant documents are displayed first. - View the knowledge base synchronization logs, verify that daily incremental synchronization tasks run normally, ensure data updates align with business rhythms.
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.