What the data for this category looks like
Core data for this category originates from the Chinese Center for Disease Control and Prevention Information System, medical institution HIS systems, infectious disease reporting systems, and laboratory testing systems. The update frequency is daily. Daily documents include infectious disease reporting cards, epidemic analysis reports, case correction records, clustered epidemic reports, and epidemiological investigation reports. Structured documents contain legally required reporting fields such as case number, diagnosis date, reporting unit, and correction flag. Unstructured documents include epidemiological investigation details and epidemic judgment content, with all fields compliant with national legal infectious disease reporting specifications.
What constraints do these characteristics impose on the vector models and indexes link?
Daily updated data requires indexes to support incremental synchronization, avoiding resource consumption and time delays caused by full reindexing. The coexistence of multiple document types requires a chunking strategy that balances precise matching of structured fields and contextual association of unstructured text. The requirement for accuracy of legally reported content means the retrieval link must strictly control false recall rates, to avoid incorrect association between policy documents and case data. The need for unified mapping of multi-source data requires index configurations to support field standardization, preventing retrieval confusion caused by data format differences across systems.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the core information length of a single infectious disease reporting card and epidemiological investigation report, avoiding splitting of key content |
| Overlap Length | 100–200 characters | Preserves cross-block associated information such as case corrections and reporting time limits, improving retrieval accuracy |
| Similarity Threshold | 0.75–0.85, calibrated via actual testing | Meets the precise matching requirement for policy documents and case data, reducing false recalls |
| Single Search Limit | Top 8 results | Covers the common associated retrieval scope for legal policy basis and monitoring data, controlling result redundancy |
defaultToken | 4096 | Adapts to the context length requirement for routine policy Q&A |
maxToken | 8192 | Supports complete retrieval and Q&A for long-term epidemic analysis reports |
Three common configuration errors
- A 503 error is returned when calling the vector model, prompting that no available channels are available under the current default group. This occurs when no dedicated channel group is configured for the text-embedding-v3 model, or when channel quotas are exhausted.
- When adjusting index parameters in the retrieval test interface, the corresponding entry cannot be found. In FastGPT v4.17.0, the index debugging entry for the retrieval test interface has been moved to the knowledge base content management page, and the operation path has not been updated synchronously.
- A large number of duplicate case reporting card entries appear after index construction. No deduplication rule configured for the case number field leads to duplicate indexing of correction records updated on the same day.
How to confirm the configuration is complete
- Initiate a policy Q&A about "legal infectious disease reporting time limits", and check whether the returned results include matching case report-related content.
- Upload a new clustered epidemic report, confirm that the index synchronization status is normal, and that the core information of this report can be retrieved in search results.
- Check the chunk length and overlap length configurations, confirm that no key policy documents or case fields are split across individual chunks.
- Verify the values of
defaultTokenandmaxToken, ensuring they cover the context requirements for routine policy Q&A and long-term epidemic analysis reports.
The values listed are common starting points, and should be evaluated against the actual samples of the target deployment.
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.