What Data for This Category Looks Like
This category of data originates from the national medical insurance information platform, hospital HIS and medical insurance interfaces, medical insurance catalog coding database, and cross-regional medical settlement platform. The update frequency is daily. Daily processed documents include medical insurance settlement lists, expense details, medical record homepages, prescriptions, rejection lists, and settlement reconciliation statements. Data fields include diagnosis and treatment project codes, drug and consumable codes, expense amounts, insured person information, settlement status, and other fields. Each document has a fixed structure. The core associated field is the correspondence between medical service items and expenses.
Constraints Imposed by These Characteristics on Knowledge Base Retrieval and Recall
Daily updated data requires the knowledge base to support incremental synchronization, avoiding the time cost of full reconstruction. Core associated coding fields require precise matching. Retrieval must prioritize associating diagnosis and treatment project and drug/consumable codes, while also accounting for consistency review of unstructured text. In bulk settlement audit scenarios, a single retrieval must cover multiple associated documents. This imposes clear constraints on recall efficiency and the number of results. Errors may lead to fund disputes, delayed hospital payments, or blocked insured person reimbursements. As such, the confidence of retrieval results must meet the strict requirements of business audits, and irrelevant content should not be returned.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets the precise matching requirements for medical insurance codes and text, balances recall precision and coverage |
limit | Top 10 entries | Adapts to the number of associated documents for a single audit, avoids overly long context that impacts LLM processing |
searchMode | vector_search | Prioritizes vector matching for codes and text content, meets precise retrieval requirements |
usingReRank | Enabled | Performs secondary sorting on recall results, filters low-relevance content, and improves audit accuracy |
datasetSearchUsingExtensionQuery | Enabled | Based on the party building learning Q&A business scenario, supplements associated policy and process text to enrich retrieval context |
collectionIds | Bind the exclusive document collection ID for medical insurance settlement audit | Isolates data from other categories, avoids interference from non-business content on retrieval results |
Three Common Misconfigurations
- When calling the knowledge base interface, a prompt indicates training order creation failed, or index update cannot be triggered. The cause is failure to distinguish between collection data addition and training order usage scenarios. Training orders are used for batch full or incremental synchronization of structured medical insurance settlement documents. Collection data addition only applies to single scattered text uploads.
- Retrieval results only return top 1 content, and cannot call top 2 or higher results for processing. The cause is incorrect configuration of the
limitparameter, or failure to specify extracting the corresponding number of retrieval results during the LLM call phase. - Attempting to upload images to the knowledge base but failing to retrieve associated content. The cause is failure to enable image vectorization configuration, or the current knowledge base not supporting extended image retrieval capabilities, making it impossible to associate image features with medical insurance business text.
How to Verify Successful Configuration
- Check the knowledge base collection's update log to confirm the daily incremental synchronization task triggers normally, matching the required data update frequency.
- Enter medical insurance codes or expense detail keywords for retrieval, and verify that the number and sorting of recall results match the configured parameter settings.
- Test calling the LLM to process retrieval results, confirming that the specified number of recall contents can be obtained to meet the context requirements of audit business.
- Try uploading image test files, confirming that the retrieval link can associate image features with medical insurance business text, and verifying that the extended retrieval function is active.
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.