What this category of data looks like
Data for this category mainly comes from the national land spatial planning one-map, project approval and supervision platform, current land use database, and ecological protection red line data. Update frequency is irregular. Daily business documents include land pre-examination and site selection opinions, construction land planning permits, construction engineering planning permits, red line drawings, and planning conditions. Core fields cover project site selection scope, land scale, planning control requirements, prior approval conditions, and more. The data structure is strongly bound to the project approval process. Associated verification relationships exist between different documents.
Constraints on knowledge base retrieval and recall
This category’s data comes from multiple dispersed sources and connects to multiple departmental systems. Retrieval must precisely match core fields such as project site selection and land scale to avoid cross-system data deviations. Data updates have no fixed cycle. Configure an incremental synchronization mechanism to ensure recalled content aligns with the latest control requirements. Document types include text and drawing formats. Support parsing and retrieval for multi-format data. Core business logic links to project approval and construction start approval. The accuracy of recalled results directly affects approval compliance. Strictly control the false recall rate.
How to set the configuration
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | This category has high precision requirements for core field matching. Filter low-correlation results to avoid errors in approval basis |
limit | Top 8 results | This category of data links to multiple prior conditions. Retain sufficient candidate results to cover full-process control requirements, while avoiding context overload |
searchMode | vector + keyword | This category has both structured fields and unstructured text. Hybrid retrieval balances precise matching and semantic recall |
usingReRank | Enabled | This category of data has strong inter-field connections. Reranking optimizes result sorting and improves the recall priority of core control requirements |
datasetSearchUsingExtensionQuery | Enabled | Business questions for this category often include prior approval related content. Extended queries cover cross-document associated information |
collectionIds | Bind to the dedicated dataset for administrative land pre-examination and planning permit | Isolate this category’s data from other natural resource-related data to avoid interference from irrelevant content in retrieval results |
Three common mistakes
- Phenomenon: Expected land red line related results are not returned during formal invocation, but work normally in debug preview. Cause: The
collectionIdsparameter was not passed in the invocation interface. No dedicated dataset is bound, and only the default general dataset is used for retrieval. - Phenomenon: A large number of non-planning control related irrelevant documents appear in retrieval results. Cause: The
similaritythreshold was not set or was set too low. Low-correlation content is recalled. - Phenomenon: Returned reference files do not match the current project's approval stage. Cause:
datasetSearchUsingExtensionQuerywas not enabled. No extended query covers prior approval related content, so only single-document content is recalled.
How to confirm the configuration is correct
- Check if the
collectionIdsparameter is passed when invoking the interface. Ensure the dedicated dataset is bound. - View the
similarityscore of retrieval results. Confirm the score falls within the preset interval. - Verify that returned reference files include core document types of this category, such as red line drawings and planning conditions.
- After enabling
usingReRank, compare result sorting before and after reranking. Confirm the priority of core control fields has improved.
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.