What Data This Category Uses
Data for this category originates from the medical insurance intelligent monitoring system, fund supervision rule base, medical insurance settlement data, drug and consumable traceability codes, and complaint clues. Daily generated documents include audit notices, inspection work drafts, suspicious item lists, interview records, handling decision letters, and rectification reports. Most individual documents include fields for designated institution names, violation descriptions, and involved amounts. No unified fixed units apply to field values; values change based on actual medical insurance settlement data. Data updates occur once per month. Total data volume grows gradually as the supervision scope expands. Document structure primarily combines structured tables and semi-structured text.
Constraints on Knowledge Base Retrieval and Recall
Monthly updated data sources require the knowledge base to support full regular synchronization and incremental updates, to avoid data lag affecting audit accuracy. The mixed structure of semi-structured and structured documents requires the retrieval link to adapt to field matching for multi-format data, rather than relying solely on pure text semantic recall. High-volume data from settlement data and traceability codes requires controlling the number of returned entries per retrieval, to avoid processing overload. The audit scenario has extremely high precision requirements: retrieval results must strictly match clauses in the supervision rule base, otherwise deviations will occur in subsequent audit processes.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Medical insurance audit requires precise matching of violation descriptions and rule base clauses. A threshold that is too low will introduce irrelevant results, while a threshold that is too high may miss potential suspicious items. Calibrated based on actual testing |
limit | Top 8–12 entries | Too many results returned in a single retrieval will exceed the context window. Too few results will fail to cover all relevant suspicious item lists and handling cases. Adapted to the multi-associated document requirements of audit scenarios |
searchMode | vector + keyword hybrid | Balances semantic recall and structured field matching, adapted to the mixed retrieval needs of structured data sources such as settlement data and traceability codes, as well as rule base text |
usingReRank | Enabled | Performs secondary sorting on initially recalled results to filter low-relevance entries, improving result precision for audit scenarios, especially for high-volume settlement data retrieval |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands user-input audit questions by adding keywords related to supervision rules, making up for differences between natural language expressions and rule base terminology |
collectionIds | Bind dedicated audit dataset ID | Only retrieves dedicated datasets containing audit notices, rule bases, and settlement data, avoiding mixing irrelevant documents from other categories and ensuring precise retrieval scope |
Three Common Misconfigurations
- Incomplete retrieval and recall, failing to cover all suspicious item lists and rule clauses. Cause: Only configured pure semantic retrieval mode, did not enable hybrid
searchMode, unable to match structured medical insurance settlement data fields and traceability code information, leading to some potential violation records not being recalled. - Workflow tool call modules cannot reference knowledge base retrieval results. Cause: Did not bind knowledge base retrieval results to workflow output fields, or did not enable result export permissions in retrieval configurations, causing workflow nodes to fail to read returned document content.
- Identical audit questions return different results. Cause: Knowledge base dataset was not locked to the current version, full historical data was not synchronized during monthly updates, or the
similaritythreshold was not fixed, leading to differences in matching results across retrieval batches.
How to Confirm Proper Configuration
- Enter the knowledge base configuration page in FastGPT v4.17.0, check the configuration values of parameters such as
similarity,limit, andsearchMode, confirm they match the preset plan. - Launch a test audit question, such as "What is the handling basis for over-prescription of medications", verify that the number of returned results matches the
limitconfiguration, and that results include rule base and handling decision letter content. - Trigger the knowledge base retrieval node in the workflow, check if the node output fields include the title and core content of the retrieval results, confirm results can be properly referenced.
- Launch the same test question three consecutive times, verify the similarity and entry consistency of returned results, confirm the
similaritythreshold and dataset version have been fixed.
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.