What the data for this category looks like
Medical insurance fund supervision and audit data sources include medical insurance intelligent monitoring systems, fund supervision rule bases, medical insurance settlement data, drug and consumable traceability codes, and report clues. The update frequency is once per month. Daily documents include audit notices, inspection work drafts, suspicious point lists, interview records, handling decision documents, and rectification reports. Document fields cover inspection period, inspected subject, basis for violation determination, traceability code number, and other items. The data includes two types of forms: structured rule entries and unstructured written records.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Monthly updated data requires regular synchronization of knowledge base indexes to avoid content lag that affects audit timeliness. Mixed document types require the retrieval mechanism to adapt to matching logic for both structured rules and unstructured records. Associated data from settlement and traceability systems requires retrieval to cover cross-field associated information, avoiding missing potential suspicious points. The most time-consuming suspicious point screening link requires retrieval results to focus precisely, reducing interference from invalid content. Retrieval errors directly relate to fund losses, so retrieval credibility must meet supervision requirements.
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 the precision requirements for matching medical insurance supervision rules, avoiding interference from low-relevance content on audit judgments |
limit | Top 10 entries | Balances the number of associated references for individual suspicious points, avoiding overly long contexts that exceed processing limits |
searchMode | Hybrid retrieval | Adapts to mixed data types of structured rule bases and unstructured inspection documents |
usingReRank | Enabled | Performs secondary sorting on recall results to improve the ranking priority of violation suspicious points |
datasetSearchUsingExtensionQuery | Enabled | Expands query terms based on supervision rules to cover more potentially associated settlement data and traceability records |
collectionIds | Bind exclusive collection IDs according to audit scenarios | Isolates knowledge base data from different supervision departments to avoid cross-scenario interference |
Three common misconfigurations
- Phenomenon: No response for a long time after initiating a conversation, and the interface continuously displays "Retrieving knowledge base". Cause: No reasonable retrieval limit and timeout parameters are set, leading to timeout when recalling content from massive medical insurance settlement data.
- Phenomenon: Chunk IDs in the knowledge base detail page cannot be selected or copied. Cause: Metadata interaction permissions for the dataset are not enabled, or front-end rendering restricts text selection functionality.
- Phenomenon: Duplicate audit suspicious point entries appear in retrieval results, reducing judgment efficiency. Cause: Index merging and deduplication configuration is not enabled, and
usingReRankis not turned on to perform secondary verification and deduplication on recall results.
How to confirm the configuration is complete
- Manually enter an audit suspicious point keyword, and check whether the retrieval results include corresponding document types of this category, such as inspection work drafts and handling decision documents.
- Check the number of returned retrieval results to confirm they match the value set for the
limitconfiguration. - View the execution logs of the knowledge base index merging task to confirm that the deduplication logic is active.
- Verify whether the extended query function is active, enter simplified violation keywords, and check whether associated supervision rule content is recalled.
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.