What data falls under this category?
Data for this category originates from online monitoring platforms, automatic pollution source monitoring systems, ecological environment law enforcement systems, and third-party test reports. Data updates occur daily. Daily documents include on-site inspection records, sampling records, monitoring reports, order for correction decisions, and administrative penalty decisions. On-site inspection records include law enforcement personnel, inspection points, and working condition information. Monitoring reports include sampling points, pollutant detection values, and detection methods. Law enforcement documents include structured and unstructured content such as illegal fact findings and rectification requirements. Data is organized and stored by law enforcement workflow, covering all stages from on-site law enforcement to rectification tracking.
What constraints do these characteristics impose on the knowledge base retrieval and recall stage?
Multi-source heterogeneous data updated daily requires retrieval to support incremental synchronization and incremental recall. This prevents full data pulls from consuming excessive system resources. Diverse document types include both structured monitoring values and unstructured law enforcement documents. Retrieval must balance accuracy for semantic matching and keyword matching. Law enforcement scenarios require precise correlation between on-site working conditions, monitoring data, and law enforcement documents. Retrieval results must cover core evidence supporting workflow decisions. Irrelevant documents must be isolated to avoid interfering with law enforcement personnel’s workflow guidance judgments.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Law enforcement scenarios require precise matching of core fields in law enforcement documents and monitoring data. This filters low-relevance historical documents to avoid interfering with workflow judgment |
limit | Top 8–12 entries | Law enforcement process guidance covers documents from multiple stages including on-site inspection, document preparation, and rectification tracking. Too many results exceed the context window, while too few fail to cover required evidence |
searchMode | Hybrid vector retrieval + keyword retrieval | Data includes structured monitoring values and unstructured law enforcement documents. Hybrid retrieval matches both semantic relevance and keyword fields, improving recall coverage |
usingReRank | Enabled | Law enforcement scenarios have high requirements for result sorting accuracy. Re-ranking re-sorts results based on semantic relevance, prioritizing recall of latest law enforcement and monitoring data |
datasetSearchUsingExtensionQuery | Enabled | Law enforcement scenarios use many business terms. Extended queries supplement synonym expressions, such as expanding "exceed standard" to "pollutant concentration exceeds limit", improving recall coverage |
collectionIds | Bind two dedicated collections: law enforcement case files and monitoring reports | Non-law enforcement knowledge base data must be isolated. This ensures retrieval results only come from dedicated documents for environmental law enforcement and monitoring, avoiding irrelevant content interfering with workflow guidance |
Three Common Misconfigurations
- Workflow and knowledge base appear blank after deployment: The interface shows empty knowledge base collections, and APIs return empty retrieval results. The cause is failure to configure persistent storage. Docker-compose deployed temporary volumes lose data after container restarts.
- Fixed prefix text appears in responses: The reply starts with a preset system guidance statement. The cause is failure to disable the system default reply prefix configuration.
- Slow retrieval speed: A single retrieval takes more than 10 seconds. The cause is failure to limit retrieval collection scope or enable re-ranking, leading to full knowledge base retrieval consuming excessive resources.
How to Verify Proper Configuration
- Check the knowledge base collection binding status. Confirm that the collections specified in the
collectionIdsconfiguration are correctly associated with the dedicated document library for environmental law enforcement and monitoring. - Submit a simulated law enforcement query, such as "How to combine sampling records and online monitoring data to judge illegal facts". Verify that the
similarityscores of returned results fall within the configured range. - Test retrieval speed to confirm single retrieval time meets business requirements. Verify that
usingReRankis active, and results are sorted to prioritize latest data. - Check system reply content to confirm no preset fixed prefix text appears. Verify that
datasetSearchUsingExtensionQuerycorrectly triggers business term expansion.
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.