What the Data for This Category Looks Like
Daily documents for this category include audit notices, inspection work drafts, suspicious item lists, interview records, handling decision letters, rectification reports, etc. Data sources cover medical insurance intelligent monitoring systems, fund supervision rule bases, medical insurance settlement data, drug and consumable traceability codes, and complaint clues. Data is updated on a monthly basis. The dataset includes structured supervision documents, semi-structured settlement details, and unstructured complaint texts. The overall data volume changes dynamically based on monthly audit task volume.
Constraints on Knowledge Base Retrieval and Recall
The monthly update rhythm requires the knowledge base to only perform full monthly synchronization, without frequent incremental updates. This reduces real-time pressure on the retrieval pipeline. The coverage of multiple document types — structured audit documents, semi-structured settlement data, and rule texts — requires the retrieval module to support accurate recall of cross-format content. Data sources cover multiple systems, so dataset permissions must be divided by business scenario during retrieval. This prevents cross-subject data leaks. At the same time, violation determination results directly relate to fund losses, and the audit link has time pressure for screening suspicious items. The relevance ranking of retrieval results must align more closely with business logic, prioritizing returns of suspicious item lists and supervision rule content.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets the rigor requirements of medical insurance supervision rules, filters out low-relevance non-audit content |
limit | Top 10 entries | Balances information density and response speed for single-round retrieval, aligns with the information needs of a single query in audit scenarios |
searchMode | Semantic retrieval + keyword hybrid | Covers precise keyword matching for rule texts and semantic association recall for settlement data |
usingReRank | Enabled | Corrects ranking deviations from initial retrieval, prioritizes high-relevance suspicious item lists and handling decision letter results |
datasetSearchUsingExtensionQuery | Enabled | Expands terms based on medical insurance supervision scenario terminology, supplements associated terms in the rule base |
collectionIds | Bind corresponding datasets by audit scenario | Isolates supervision data from different subjects, complies with government data security requirements |
Three Common Configuration Mistakes
- Phenomenon: Retrieval response time exceeds expectations, and the interface displays a loading timeout. Cause: The enabled status of
limitandusingReRankis not adjusted based on hardware configuration. Vector recall and reranking tasks occupy too much computing power, exceeding the processing limits of GPU video memory and CPU cores. - Phenomenon: Retrieval results do not change after enabling
usingReRank. Cause: No pre-filtering threshold forsimilarityis set. Initial retrieval only returns high-similarity results, and the reranking module has insufficient data for ranking optimization. - Phenomenon: Suspicious item lists do not appear in priority positions in retrieval results. Cause: Only the semantic retrieval mode of
searchModeis enabled. Keyword matching for violation keywords in audit documents is not combined, so the ranking logic does not fit the business scenario.
How to Verify Correct Configuration
- Enter the knowledge base configuration page of FastGPT v4.17.0, and verify that the values of core configuration items such as
similarity,limit, andusingReRankmatch the preset plan. - Upload a single audit notice and medical insurance settlement data sample, perform a test retrieval, and check whether the returned results prioritize displaying suspicious item lists and supervision rule texts.
- Trigger the monthly knowledge base synchronization task, verify that the task can be completed within 24 hours, and there are no data synchronization errors.
- View the retrieval log, confirm that
datasetSearchUsingExtensionQueryhas triggered scenario-based terminology expansion, and the returned results include associated fund supervision rules.
The values provided are common starting points and should be evaluated using local test 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.