What this type of data looks like
This type of data primarily comes from the territorial and spatial basic information platform, natural resource survey and monitoring data, population and economic statistics data, and remote sensing images, with an annual update frequency. Daily processed documents include territorial and spatial overall plans, detailed plans, special plans, planning implementation evaluation reports, and public opinion summaries. Data fields include spatial coordinate ranges, constrained indicator values, department and public opinion entries, and other fields. Some spatial data requires matching a dedicated projection coordinate system, and there is no unified fixed general unit.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The annual update feature means the knowledge base only needs to complete full data synchronization on a fixed cycle, without frequent incremental updates, which reduces daily maintenance costs. The multi-source heterogeneous data structure, including spatial coordinates, statistical values, and textual opinions, requires the retrieval link to support cross-field matching, while also distinguishing recall rules for spatial data and non-spatial data to avoid result deviations caused by mixed matching. Document types cover planning texts and evaluation reports, requiring recall weight configuration by document category to avoid irrelevant documents interfering with the accuracy of ledger summary. Regional-level planning data has a large volume, so the number of single-recall results must be limited to prevent context overflow that affects the coherence of report generation.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Territorial spatial planning data contains a large number of professional terms and spatial constraints, requiring filtering of low-match irrelevant results to ensure the accuracy of ledger summary |
limit | Top 10 entries | Regional planning data has a large volume, and too many recall results will exceed the context window and affect the coherence of report generation |
searchMode | vector+keyword hybrid retrieval | The data includes structured fields such as spatial coordinates and professional textual descriptions of planning; hybrid retrieval balances semantic matching and precise field matching |
usingReRank | Enabled | Planning documents have strong professionalism, and reranking optimizes the ranking of high-match results, avoiding professional terms being crowded out by low-match documents |
datasetSearchUsingExtensionQuery | Enabled | Planning questions often include professional abbreviations or specific spatial scope terms; extended queries can cover more relevant documents |
collectionIds | Bind dedicated territorial spatial planning dataset ID | Business data from other categories must be isolated to avoid recalling irrelevant content |
Three common configuration errors
- Symptom: The knowledge base question answering interface does not display recommended question buttons matching planning scenarios. Cause:
datasetSearchUsingExtensionQueryis not configured to generate scenario-based associated questions, so the recommendation logic cannot be triggered. - Symptom: Non-planning documents are mixed in the retrieval results. Cause:
collectionIdsis not configured to bind the dedicated dataset, leading to recall of business data from other categories. - Symptom: The generated ledger fails to cover the basis of the specified planning scope. Cause: The value of
limitis set too low andusingReRankis not enabled, so valid results are filtered out, resulting in insufficient information.
How to confirm the configuration is complete
- Log in to the FastGPT v4.17.0 knowledge base management backend, check whether
collectionIdshas been bound to the dedicated territorial spatial planning dataset ID. - Initiate a test query containing planning scope and constrained indicators, verify whether the
similarityscore of the returned results falls within the preset 0.75–0.85 range. - View the retrieval result list, confirm that the number of results matches the top 10 setting specified by
limit, and no documents from other categories are mixed in. - After enabling
usingReRank, initiate the test query again, confirm that professional planning documents are ranked higher than low-match 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.