Reference only; verify against the target FastGPT release and deployment documentation.
What Data for This Category Looks Like
This category of data mainly comes from the National Center for Disease Control and Prevention Information System, hospital HIS systems, infectious disease reporting systems, and laboratory testing systems. It updates daily. Commonly generated documents include infectious disease report cards, epidemic analysis reports, case correction records, clustered epidemic reports, and epidemiological investigation reports. These documents contain core fields such as case identifiers, diagnostic information, reporting time limits, and correction status. Individual document content ranges from hundreds to thousands of characters. Associated fields required by legal infectious disease reporting rules must be strictly matched.
Constraints Imposed on Knowledge Base Retrieval and Recall by These Characteristics
Data sources for this category are scattered, and multi-system field association is required. Retrieval must accurately match associated information such as case identifiers across documents. The daily high-frequency update feature requires the retrieval pipeline to support fast data synchronization. This avoids response bias caused by stale data. Documents include dynamically corrected content. Retrieval must distinguish the priority of original reports and correction records to ensure that the latest valid information is returned. Field association logic required by legal reporting rules means retrieval cannot rely solely on keyword matching. It must balance field completeness and compliance. Individual document content varies widely. Proper control of retrieved block length is needed to avoid context redundancy that affects retrieval efficiency.
Configuration Recommendations
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Must match core fields such as precise case identifiers and diagnostic terms, filter low-relevance content |
limit | Top 8–12 entries | Cover associated documents from multiple sources, while avoiding excessive content exceeding the context processing window |
searchMode | Hybrid Retrieval | Balance keyword matching and semantic understanding, adapt to retrieval needs of different document types such as report cards and epidemiological investigation reports |
usingReRank | Enabled | Correction records and original reports have high semantic similarity, re-ranking can prioritize returning latest valid content |
datasetSearchUsingExtensionQuery | Enabled | Expand search terms based on core case fields to cover associated data across systems |
collectionIds | Configured by business scenario grouping | Distinguish knowledge base collections for different infectious disease categories or reporting cycles, accurately limit retrieval scope |
Three Common Misconfigurations
- Phenomenon: The number of retrieval results does not reach the configured
limitupper limit, and some expected case correction records are not retrieved. Cause: Thesimilaritythreshold is set too high. This filters correction records with semantic relevance but slightly lower keyword matching degree.usingReRankis not enabled to re-rank similar results. - Phenomenon: Multiple duplicate report records for the same case appear in retrieval results. Cause: Dedicated retrieval grouping is not configured for different document types. Multiple indexes of the same document block are retrieved at the same time, and effective deduplication is not performed.
- Phenomenon: The context content returned by the retrieval request exceeds the system processing limit, causing response failure. Cause: The
limitparameter is not set properly. Too many document blocks are retrieved. Low-relevance content is not filtered via thesimilaritythreshold, resulting in total character count exceeding the upper limit.
How to Confirm Proper Configuration
- Input a test query containing core case identifiers. Verify that returned results include the latest valid documents for the corresponding case.
- Confirm that the retrieval scope is limited to the knowledge base grouping for the current business. This avoids retrieval of irrelevant content.
- View retrieval pipeline logs. Confirm that the number of retrieved results meets the preset retrieval upper limit requirements.
- Compare retrieval results with and without the re-ranking function enabled. Confirm that the re-ranking logic prioritizes returning latest correction content.
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.