What data for this category looks like
Data for this category primarily originates from meteorological warning systems, hydrological monitoring systems, emergency command platforms, grassroots disaster reporting systems, and video surveillance. The update frequency is daily. Daily documents include duty logs, warning notices, consultation minutes, rapid disaster reports, emergency response plans, and material allocation slips. Some documents contain fields such as real-time monitoring points, water level values, and personnel transfer ledgers. Data exists in a mixed format of structured and unstructured content. Some associated video materials require synchronized associated retrieval.
Constraints imposed on knowledge base retrieval and recall by these characteristics
Scattered data sources and inclusion of real-time monitoring data require retrieval to cover synchronized data across multiple systems, to avoid information silos. The daily update feature requires the retrieval pipeline to adapt to high-frequency incremental synchronization of knowledge bases, to prevent recalling outdated content. Diverse document types and numerical fields such as water levels and monitoring points require mixed retrieval of structured data and unstructured text. The presence of associated video materials requires the retrieval pipeline to support cross-modal recall, and to match context information of corresponding monitoring points to ensure retrieval results are strongly linked to on-site scenarios.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Matches the semantic matching accuracy required for flood prevention warning scenarios, and avoids recalling irrelevant historical disaster records |
limit | Top 12 entries | Covers multiple types of documents required for a single consultation, while controlling context length to stay within model limits |
searchMode | hybrid | Combines keyword and semantic retrieval to meet mixed retrieval needs for numerical fields and text |
usingReRank | Enabled | Performs secondary ranking on initial recall results to prioritize content associated with current rainfall and water conditions |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval terms to cover dialectical disaster descriptions in grassroots disaster reports |
collectionIds | Configured by business group | Limits retrieval scope to the exclusive knowledge base for emergency duty, avoiding mixing in data from other businesses |
Three Common Configuration Errors
- Phenomenon: A
timeout of 60000ms exceedederror occurs when deleting folders with a large number of documents in the knowledge base. Cause: No segmented processing logic is configured for large-volume batch document operations, resulting in single-batch operation timeout. - Phenomenon: Images associated with some retrieval results cannot be displayed normally. Cause: No cross-monitoring system image resource path mapping is configured, causing retrieval results to fail to load original stored image files.
- Phenomenon: Retrieval results do not match structured rainfall and water level numerical content. Cause: Mixed retrieval mode is not enabled; relying solely on semantic retrieval cannot cover matching requirements for numerical fields.
How to Verify Successful Configuration
- Initiate a retrieval containing scenario keywords such as "rainfall", "water level", "emergency response", and check whether the returned results cover corresponding types of documents and structured fields.
- Enter the knowledge base retrieval configuration panel, confirm that the switch states of
usingReRank,datasetSearchUsingExtensionQueryand other items match the preset configuration. - Test deleting a folder containing more than 100 documents, confirm that no timeout error is triggered during the operation.
- Upload a test document containing monitoring point images, initiate associated retrieval, and check whether the image links can be accessed normally.
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.