Declaration Materials and File Review for Territorial Spatial Planning Compilation and Implementation Evaluation: Knowledge Base Retrieval and Recall

What data for this category looks like

Daily documents for this category include territorial spatial overall plans, detailed plans, special plans, planning implementation evaluation reports, public opinion summaries, and more. Data sources cover territorial spatial basic information platforms, natural resource survey and monitoring data, population and economic statistical data, and remote sensing image materials. Data is updated on an annual basis. Document structures include structured control fields such as spatial coordinates, controlled area (hectares), planning period (years) and unstructured text content. Some documents include associated spatial layer information.

What constraints these characteristics impose on knowledge base retrieval and recall

Multi-source heterogeneous data sources require the retrieval system to support both text semantic matching and spatial attribute association, to avoid missing cross-data-source associated content. Annual update frequency means the knowledge base incremental synchronization cycle does not need to be overly short, and can be updated in annual batches. Document volumes vary widely, from several hundred thousand-word overall plan texts to several hundred-word public opinion summaries. This requires retrieval configurations to balance long text chunking and accurate short document recall. Business error consequences are directly linked to planning approval and project implementation. Recall result accuracy and relevance must meet review standards, and low-match or irrelevant content must not be included.

How to set configurations

Reference only; verify against the target FastGPT release and deployment documentation.

Three common errors

• Symptom: Locally uploaded planning documents cannot be retrieved during conversations, but files uploaded in bulk via the knowledge base backend can be retrieved normally. Cause: File parsing adaptation rules for conversation scenarios are not configured, and only the parsing process for knowledge base bulk uploads is enabled.
• Symptom: Knowledge base retrieval tests work normally, but empty results are returned during conversations. Cause: datasetSearchUsingExtensionQuery is not enabled, or the similarity threshold is set too high, causing professional term expansion to fail or filtering eligible relevant content.
• Symptom: Retrieval results include non-planning spatial data or statistical content. Cause: collectionIds is not bound to an exclusive business collection, or searchMode is not limited to the planning business domain retrieval mode.

How to confirm configurations are properly set

• Upload a single territorial spatial overall plan document to a test knowledge base, perform keyword retrieval, and verify whether recall results include the document's core control indicators and planning content.
• Adjust the similarity parameter range, compare the number of recall results across different thresholds, and select a value range that meets business review requirements.
• After enabling usingReRank, verify whether the sorting priority of multi-source data such as remote sensing images and statistical data aligns with planning review business logic.
• Test knowledge base access permissions using different role accounts, confirm that collections bound via collectionIds are only accessible to authorized business roles.

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.