What Data for This Category Looks Like
Data for this category comes from the national medical security information platform, hospital HIS and medical insurance interfaces, medical insurance catalog and coding library, and cross-provincial medical settlement platform. Daily processed documents include medical insurance settlement lists, expense details, medical record homepages, prescriptions, rejection payment lists, and settlement reconciliation statements. Single documents usually contain fields such as diagnosis and treatment item codes, expense amounts, and insured person identity identifiers. The update frequency is daily, and coverage is required for single or batch medical expense payment-related documents and materials.
What Constraints Do These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
Daily data updates require the knowledge base to be configured with an automatic synchronization mechanism. This ensures core data such as medical insurance catalogs and settlement rules are available in real time, preventing use of expired standards during audits. Documents include strongly verified fields like diagnosis and treatment item, drug, and consumable codes. Retrieval must accurately match these codes to avoid expense audit errors. Structural differences between single or batch settlement documents require retrieval to support filtering by document type. It also requires controlling the format of recalled results to adapt to document drafting needs of different audit scenarios.
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 strongly verified fields such as medical insurance codes. The similarity threshold must be sufficient to filter irrelevant results while covering compliant code variants. |
limit | Top 8–12 entries | Settlement audits require reference to multiple related documents such as prescriptions, settlement lists, and medical insurance catalogs. 8–12 entries cover common related scenarios. |
searchMode | Exact match first | Adapts to business scenarios requiring precise matching such as medical insurance codes and expense rules, avoiding generalized recall that leads to audit errors. |
usingReRank | Enabled | Reorders recalled results by relevance to improve the display priority of high-match medical insurance catalogs and settlement rule documents. |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms, adding synonyms and common aliases for medical insurance codes to adapt to differences in code expressions reported by different channels. |
collectionIds | Bind dedicated collections for medical insurance catalogs, settlement rules, and local settlement documents | Isolates business data to avoid interference from irrelevant knowledge base content on settlement audit retrieval results. |
Three Common Misconfigurations
- Long documents such as monthly settlement reconciliation statements are incorrectly split into meaningless fragments, preventing retrieval from matching complete expense details. The cause is failure to adjust chunking rules for the field structure of medical insurance documents, using general chunking parameters.
- Retrieval results return only 2 or fewer entries, unable to cover related content of prescriptions and settlement lists. The cause is failure to adjust the
limitparameter, using the default low recall count configuration. - Retrieval results include generalized content unrelated to medical insurance codes, leading to audit errors. The cause is failure to set
searchModeto exact match first, enabling generalized retrieval mode.
How to Verify Proper Configuration
- Manually upload a medical insurance settlement list document, and check if the corresponding medical insurance catalog code data can be recalled during retrieval.
- Adjust the
similarityparameter, and verify if the match quality of recalled results meets business verification requirements. - Trigger an automatic synchronization task, and check if the update time of knowledge base data matches the source data update time.
- Switch
searchModeto exact match mode, and verify if the retrieval accuracy of code fields has improved.
The values provided 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.