What the data for this category looks like
This category's data mainly comes from the National Medical Security Information Platform, hospital HIS and medical insurance interfaces, medical insurance catalog and coding database, and cross-regional medical settlement platform. The update frequency is daily. Daily business documents include medical insurance settlement lists, expense details, medical record homepages, prescriptions, rejection payment lists, and settlement reconciliation statements. They cover fields such as diagnosis and treatment item codes, drug and consumable codes, expense amounts, and insured person information. Some documents contain structured tables and image content, which are used for expense matching and audit verification.
What constraints these characteristics impose on the "knowledge base retrieval and recall" link
Daily updated data sources require the retrieval process to support incremental data synchronization. This prevents retrieval results from lagging behind the latest medical insurance policies and settlement data. Multi-type document structures including tables and images require retrieval tools to support structured content extraction. Without this support, core audit fields such as codes and details will be lost. Fixed business fields such as codes and expense amounts require retrieval to balance precise matching and semantic association. This avoids introducing irrelevant data due to fuzzy matching. Business scenarios of single or batch settlement audit require retrieval to support single requests and batch requests. It also requires controlling the amount of returned information to avoid interfering with audit efficiency.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Medical insurance data fields are precise. Low-match irrelevant results must be filtered to ensure the validity of audit basis |
limit | Top 10 entries | Settlement audit needs to balance multi-dimensional associated data. Too many returned results increase manual verification costs. Too few results may miss key information |
searchMode | vector+keyword | This covers precise matching of structured coding fields and semantic retrieval of unstructured text. It adapts to multiple types of business documents |
usingReRank | Enabled | Precise matching of the association between diagnosis and treatment items and expenses is required. Reranking optimizes result sorting priority and improves audit efficiency |
datasetSearchUsingExtensionQuery | Disabled | Medical insurance data fields are fixed and business logic is clear. No additional query generation for redundant content is needed |
collectionIds | Bind specified medical insurance knowledge base collections | Non-medical insurance business data must be isolated. This ensures the retrieval scope only covers compliant medical insurance settlement-related documents |
Three common mistakes
- Table and image content in retrieved documents cannot be displayed normally. The cause is that no dedicated extraction rules are configured for structured tables and images in medical insurance settlement lists and rejection payment lists. Only ordinary text chunking is used, resulting in loss of core structured information.
- The retrieval knowledge base scope cannot be switched in real time during conversations. The cause is that no dynamic collection binding logic is configured. Only the fixed
collectionIdsparameter is supported, so the data source collection for retrieval cannot be adjusted during a session. - Recommended "questions you may ask" content does not match actual questions in the knowledge base. The cause is that automatic extended query generation is enabled for recommendations. Real questions from the knowledge base are not used as the recommendation source, leading to recommended content that does not match the business scenario.
How to confirm the configuration is correct
- Upload a medical insurance settlement list document, perform retrieval and verification, and confirm that returned results include core fields such as codes and expense details from the document.
- Adjust the
similarityparameter range, and verify whether the matching accuracy of retrieval results meets business audit requirements. - Test the batch retrieval scenario, and confirm that the number of returned results does not exceed the range set by the
limitparameter. - Check whether
usingReRankis enabled, and verify whether reranked results prioritize matching the association logic between diagnosis and treatment items and expenses.
The values provided here are common starting points and should be evaluated against your own 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 2026-09-28.