What this category of data looks like
The typical stakeholders for this category are natural resources department planning divisions, planning institutes, local governments, and industry regulatory authorities. Daily documents include territorial spatial master plans, detailed plans, special plans, planning implementation evaluation reports, and public opinion summaries. Data sources cover territorial spatial basic information platforms, natural resources survey and monitoring data, population and economic statistical data, and remote sensing imagery. Update frequency is once per year. Most documents contain fields such as spatial coordinates, constraint indicators, and department coordination opinions. Some documents include vector layer metadata. Industry standard units such as hectares, square kilometers, and population counts are used.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
The annual fixed update frequency requires the knowledge base to perform full synchronization on a regular basis. It prevents the index from being out of sync with the latest policies and spatial data. Multi-source heterogeneous data sources require compatibility with vector files, spreadsheets, and unstructured documents. The retrieval link must balance matching of text semantics and spatial correlation information. Professional terminology and spatial fields in documents require the retrieval model to have industry semantic recognition capabilities. This avoids misjudgment of terms. Unstructured content in public opinion summaries and department coordination documents accounts for a relatively high proportion. Chunking strategies need to be optimized to retain contextual associations. This prevents key policy basis from being split and lost.
How to configure
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the rigor of territorial spatial planning policy documents, filters low-relevance non-policy content |
limit | Top 10 results | Adapts to the requirement of covering multiple planning basis for single-round Q&A, avoids excessive context load |
searchMode | Semantic retrieval | Adapts to long-paragraph semantic associations in policy texts, matches policy connection content across documents |
usingReRank | Enabled | Performs secondary ranking on initially recalled results to correct local deviations from semantic retrieval |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonyms and industry terms related to policies, improves recall coverage of policy basis |
collectionIds | Bind exclusive planning knowledge base collections | Isolates planning data from other business data to ensure precise retrieval scope |
Three common mistakes
- Non-planning documents returned in search results: The cause is failure to bind exclusive
collectionIds. The retrieval scope covers data from other business collections. - Batch data push failure: The cause is uploading more than 200 data groups in a single
pushDataAPI call. Data must be split into batches. - Knowledge base content accidentally exposed to the public internet: The cause is not enabling access permission restrictions for private deployment. Public network access to knowledge base collections is allowed.
How to confirm the configuration is complete
- View the binding configuration of the knowledge base collection. Confirm that
collectionIdsonly includes datasets related to territorial spatial planning. - Initiate a policy-related test query. Verify that the
similarityscore of returned results falls within the preset 0.75–0.85 range. - Call the
pushDataAPI to push test data within 200 groups. Confirm that the API returns a successful status code. - Check the access control rules for private deployment. Confirm that public network access permissions for the knowledge base are not enabled.
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.