What this category of data looks like
This category of data primarily originates from the national spatial planning one-map platform, project approval and supervision platform, land use status database, and ecological protection redline data, with irregular update frequency. Daily processed documents include land pre-examination and site selection opinions, construction land planning permits, construction engineering planning permits, redline diagrams, and planning conditions. It includes fixed fields such as project number, land scale, control boundary, and approval node. The data mixes two formats: text-based approval documents and vector attribute data. The content directly links to core requirements for project approval and construction commencement approval.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Data sources are scattered and updates are irregular. Regular synchronization of incremental data is required to ensure the timeliness of retrieval content. The mixed format of text-based approval documents and vector attribute data means retrieval methods must adapt to both semantic matching and precise field matching. Fixed fields such as project number, land scale, and control boundary have strong business relevance. Retrieval must prioritize matching precise attributes related to the project to avoid interference from irrelevant planning data. Since the data directly links to project approval and construction commencement approval, recalled results must have high confidence. Low-relevance or incorrect data must be prevented from entering the approval process.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches precise semantics related to land control, filters low-relevance general planning data, and ensures business relevance of recalled results |
limit | Top 8 entries | Covers the number of core pre-submitted documents required for a single project approval, balances recall volume and context length, and avoids exceeding the model's context window |
searchMode | hybrid | Adapts to the mixed retrieval needs of text-based approval documents and vector attribute data, covering both semantic and precise field matching |
usingReRank | Enabled | Performs secondary sorting of recalled results by project relevance, corrects deviations from single vector retrieval, and improves the ranking priority of high-relevance results |
collectionIds | Bind the specified land pre-examination and planning permit knowledge base | Limits the retrieval scope, avoids interference from redundant data across knowledge bases, and implements knowledge base priority control |
datasetSearchUsingExtensionQuery | Enabled | Supplements extended retrieval for fields such as project number and land scale, matches precise business attributes, and improves recall accuracy for specific fields |
Three Common Configuration Mistakes
- Failing to configure
collectionIdsto limit the retrieval scope results in invalid cross-knowledge base recall. This occurs because the target knowledge base is not specified, so retrieval covers all mounted data sources and introduces planning data unrelated to the current project. - Failing to enable document format parsing configuration causes context references to only display plain text chunks, without rendering according to the original file format. Only original text chunk content is retained, and the original document structure is not restored.
- A 60-second timeout prompt appears when switching knowledge base indexes. This is because the
limitparameter for single retrieval is not restricted, and the data source includes uncompressed large-volume redline diagram data, causing vector retrieval time to exceed the threshold.
How to Verify Correct Configuration
- Enter the FastGPT knowledge base management interface. Confirm that
collectionIdshas been bound to the specified land pre-examination and planning permit related knowledge base, and no irrelevant data sources are included. - Initiate a test query containing keywords such as project land scale and control boundary. Check whether the
similarityscore of the returned results falls within the 0.75–0.85 range. - View the context reference module. Confirm that the document format has been rendered according to the original file style, and no pure Markdown code blocks appear.
- Simulate a multi-knowledge base retrieval scenario. Verify that only the specified target knowledge base returns valid results, and no cross-library redundant data appears.
The values provided here are common starting points and should be evaluated 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.