What Data Looks Like for This Category
The service subjects for this category include housing and construction quality and safety supervision stations, construction enterprises, supervision enterprises, and construction units. Data sources cover engineering quality and safety supervision systems, construction site real-name management platforms, supervision logs, and test reports. The update frequency is weekly. Daily business documents include supervision plans, supervision inspection records, hidden danger rectification notices, quality problem handling forms, and completion acceptance supervision reports. Most documents contain dedicated fields such as project number, construction enterprise name, hidden danger description, rectification deadline, and acceptance conclusion. Individual document lengths vary significantly. Some test reports include structured data and attached images.
Constraints Imposed on Knowledge Base Retrieval and Recall
Multi-source heterogeneous data sources require retrieval to support unified indexing of cross-system data. This avoids missed recalls caused by format differences. Weekly update cycles require configuring incremental synchronization mechanisms. These ensure knowledge base data updates synchronously with business systems, without lagging behind review needs. Documents include dedicated business fields such as project numbers and hidden danger types. Retrieval must support precise matching of specific fields to associate multiple process materials under the same project. Errors here can lead to quality or safety accidents. Recall accuracy requirements are high, so irrelevant documents must be excluded from the review scope.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets precision requirements for matching business key fields, avoids recalling irrelevant documents or missing core materials |
limit | Top 10 entries | Controls context length, adapts to the business scenario of associating multiple materials for file review |
searchMode | vector+keyword | Balances semantic matching and precise matching of structured fields such as project numbers and hidden danger types |
usingReRank | Enabled | Improves sorting priority for highly relevant documents, reduces missed detection risk |
datasetSearchUsingExtensionQuery | Disabled | Prevents extended queries from introducing general content outside the knowledge base, strictly limits retrieval scope |
collectionIds | Bind corresponding knowledge base ID by business domain | Restricts retrieval to only call dedicated data for this category, avoids mixing documents from other categories |
Three Common Configuration Mistakes
- Phenomenon: Retrieval results include general engineering knowledge outside the knowledge base, and are not strictly limited to documents for this category. Cause: Failed to correctly configure
collectionIdsto bind the dedicated knowledge base, or enableddatasetSearchUsingExtensionQuery, causing extended queries to introduce irrelevant content. - Phenomenon: When attempting to download test reports or supervision logs uploaded to the knowledge base, a 404 error occurs or the download address is incorrectly spliced, making it impossible to obtain complete files. Cause: Failed to confirm the correct call path of the "View Original Content" interface, or file storage permission configuration did not enable download permissions.
- Phenomenon: When retrieving files under the same project, recalled documents only include a single rectification notice, and the corresponding supervision inspection records are missing. Cause: The
similaritythreshold is set too high, or keyword matching insearchModeis not enabled, making it impossible to match multiple documents associated with the project number.
How to Verify Correct Configuration
- Perform a retrieval using a single project number. Check whether recalled documents only include supervision plans, rectification notices and other materials for that project.
- Enable retrieval logs. Check whether only the bound dedicated knowledge base ID is called, and no global extended query is triggered.
- Upload a new weekly updated supervision log. Wait for synchronization to complete, then perform retrieval to confirm that new data has been recalled.
- Adjust the value of the
similaritythreshold. Observe changes in the number of recalled documents, and confirm that the threshold configuration takes effect.
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.