What data looks like for this category
Data for this category comes primarily from the Chinese Center for Disease Control and Prevention Information System, medical institution HIS and infectious disease reporting systems, and laboratory testing systems. Commonly processed documents include infectious disease reporting cards, epidemic analysis reports, case correction records, clustered epidemic reports, and epidemiological investigation reports. Data updates occur daily. Individual documents include standardized fields such as basic case information, diagnosis results, reporting deadlines, and correction information. Document structures are fixed and comply with statutory reporting specifications.
What constraints these characteristics impose on vector models and indexing
The fixed structure of data for this category and its daily update rhythm require the indexing process to support rapid matching of standardized fields. Data from multiple systems may have field alignment differences. Basic field unified mapping must be completed before indexing. Because data is updated daily, indexing must support incremental updates to avoid the time cost of full reconstruction. The core review link relies on key fields such as case diagnosis and reporting deadlines. Indexing must prioritize the vector recall accuracy of these fields to ensure abnormal signals related to underreporting or delayed reporting can be quickly identified.
How to Set Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | This range covers the core review fields of a single case report, as documents for this category are a mix of structured and semi-structured content. |
| Overlap Length | 100–200 characters | Prevents critical information such as reporting deadlines and case IDs from being truncated across chunks, ensuring complete vector recall. |
Reference only; verify against the target FastGPT release and deployment documentation.
| maxToken | 4096 token | Covers the maximum content length of a single epidemiological investigation report or epidemic analysis report, avoiding truncation of core review information. |
| Similarity Threshold | 0.75–0.85 | Balances abnormal signal recall rate and false positive rate, and meets the review requirement of zero tolerance for underreporting for this category. |
| Single Recall Limit | Top 20 entries | Meets the needs of abnormal signal screening during daily summary reviews, avoiding excessive redundant results that interfere with manual verification. |
Three Common Mistakes
- Phenomenon: After calling the knowledge base interface to create a file collection, the vector model changes automatically, and previously imported documents cannot match the new vector encoding. Cause: The specified vector model was not locked during creation. The platform uses the latest updated model by default, leading to inconsistent index encoding.
- Phenomenon: After importing daily updated infectious disease reporting cards, search responses time out, and index generation time increases significantly. Cause: Incremental index mode was not configured. Full reconstruction is performed for each update, and the chunk length is set outside the reasonable range, increasing single-batch processing pressure.
- Phenomenon: When searching for case correction records, documents with low similarity appear at the top, making it impossible to quickly locate target content. Cause: Weighting was not configured for the core review fields of this category. The default sorting only relies on global vector similarity, and does not distinguish the priority of key information such as case ID and reporting deadlines.
How to Confirm Correct Configuration
- Check the vector model configuration item of the file collection to confirm it matches the model specified by the business requirements, avoiding mismatches caused by default model changes.
- Import a single test infectious disease reporting card and verify that the chunked text length falls within the preset range, with no severe truncation.
- Initiate a search for core review fields such as case ID and reporting deadlines, and confirm that the number of recall results matches the single recall limit setting.
- Perform an incremental update operation and check that only newly imported documents are added to the index, with no full reconstruction of the existing index required.
The values provided here 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.