What this category of data looks like
Data for this category comes from meteorological warning systems, hydrological monitoring systems, emergency command platforms, grassroots disaster reporting systems, and video surveillance. Daily documents include duty logs, warning notices, consultation minutes, rapid disaster reports, emergency response plans, and material allocation slips. Some documents include fields such as rainfall and water level values, material numbers, evacuation zones, and warning levels. Data updates happen daily. Some real-time warning information updates dynamically alongside monitoring data. Most documents combine structured fields and unstructured text. Some rapid disaster reports include on-site photo or video links.
Constraints Imposed on Knowledge Base Retrieval and Recall
The data is multi-source and scattered. This requires the retrieval link to support cross-system associated data recall, eliminating information silos. Data updates daily, so a near-real-time index update mechanism is needed. This ensures policy and warning information stays timely. Documents have both structured fields and unstructured text. This means the system must support both field-based precise matching and semantic retrieval. This covers queries for policy clauses, response processes, and other content. Emergency duty scenarios have strict response speed requirements. The recall link must limit the number of results returned per request and define sorting logic. This avoids redundant information interfering with response decision-making.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.70–0.85 | Matching policy and regulatory basis requires balancing accuracy and recall volume. This interval filters low-relevance content while covering associated clauses |
limit | Top 5–7 results | Emergency scenarios require quick access to core information. Limiting results avoids information overload and lets duty personnel review content quickly |
searchMode | semantic + keyword | Balances semantic understanding of policy clauses and precise keyword matching, adapting to different policy query needs |
usingReRank | Enabled, retain top 3 results after reranking | Multi-source recall results may include redundancy. Reranking prioritizes content with higher relevance to policy basis |
datasetSearchUsingExtensionQuery | Enabled, extended vocabulary size of 150–250 words | Automatically expands user policy queries into associated terms, covering more potential regulatory basis documents |
collectionIds | Specify exclusive collection IDs for flood control and disaster prevention policies | Isolates flood control and disaster prevention policy documents to avoid irrelevant data interfering with retrieval results |
Three Common Mistakes
- Issue: Retrieval results only return 1 entry, and the first 2 results cannot be selected for large model processing. Cause: The
limitparameter value range is not configured correctly. The default setting only returns a single result, and downstream result handling logic is not adjusted. - Issue: After uploading policy documents via collection data addition, new content cannot be retrieved. Cause: The responsibilities of collection data addition and creating training orders are confused. Collection data addition only completes text upload and does not trigger index construction. A training order must be created to start vectorization and index generation.
- Issue: Retrieving policy basis returns many irrelevant daily duty logs, and fails to precisely match regulatory clauses. Cause: The
collectionIdsparameter is not configured to specify the exclusive policy collection. The retrieval scope covers unrelated documents, leading to redundant results.
How to Confirm Proper Configuration
- Upload one flood control policy document, run the create training order operation, and check if the index generation status shows completed.
- Enter a query phrase including "flood control warning levels", and verify that retrieval results include the policy document and match the configured
limitresult count. - Adjust the
similarityparameter interval, and confirm that retrieval result relevance changes appropriately with the parameter. - After enabling
usingReRank, confirm that reranked results prioritize content related to policy basis.
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.