What this category of data looks like
Data for this category draws from regulatory standards, online environmental monitoring equipment, gas detectors, operator qualification files, equipment isolation lists and supervision platforms. Daily documents cover risk identification records, work permits for hot work, confined space work, lifting operations and other types, environmental monitoring logs, hazardous waste ledgers, emission monitoring reports, and accident and hidden danger rectification reports. Data updates are submitted in real time upon completion of a single operation, or synced according to monitoring cycles and regular inspection requirements. Document fields include operation time, gas concentration values, qualification validity periods, isolated equipment numbers and other items. Units include ppm, cubic meters, days and other professional measurement identifiers.
What constraints do these characteristics impose on knowledge base retrieval and recall
Multi-source and heterogeneous data sources require retrieval to cover different document formats and types. This avoids missing compliance-related content. Real-time or high-frequency updated data requires the knowledge base to support incremental synchronization. This ensures that content cited in bid documents matches the latest on-site status. Professional fields and units require retrieval to match precise terminology. This prevents semantic deviations. High-risk business attributes require recall results to have high confidence. This avoids compliance risks caused by incorrect citations. Length limits for bid documents require retrieval results to be controlled in quantity. This prevents context overflow.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirement for safety and environmental protection terminology, avoids low-relevance content from being included in bid documents |
limit | Top 8 entries | Covers multiple types of documents including work permits, qualifications, regulations and others, while controlling context length to meet the processing limit of FastGPT v4.17.0 |
searchMode | embedding | Adapts to semantic matching of professional terminology, identifies associated content for operation scenarios |
usingReRank | Enabled | Performs secondary sorting on recall results, filters low-confidence compliance content, reduces the probability of bid errors |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval scope, associates content between regulatory standards and on-site monitoring data, improves retrieval comprehensiveness |
collectionIds | Filter by EHS business tags | Only import dedicated datasets such as risk identification, work permits, hazardous waste ledgers and others, to avoid irrelevant data interfering with retrieval results |
Three common mistakes
- Retrieval results include non-EHS business documents. The cause is that the
collectionIdsfiltering rule is not configured, and knowledge base content from other categories is mixed in. - Image links stored in the knowledge base cannot be displayed in conversations. The cause is that the context transfer configuration for external image links is not enabled, or the image link is not bound to the corresponding business document field.
- When importing a PPT-format knowledge base, only text is displayed without images. The cause is that the PPT image extraction configuration is not enabled, or the image parsing process is not correctly associated with the FastGPT service.
How to confirm the configuration is correct
- Initiate a test query covering EHS operation scenarios, verify that retrieval results only include dedicated business documents such as risk identification records and work permits.
- Upload a PPT document containing images to the knowledge base, confirm that image links can be extracted and displayed normally in conversations after import.
- Adjust the
similaritythreshold, observe changes in the relevance of retrieval results, confirm that the threshold meets business precision requirements. - Trigger the retrieval reference merging function, verify that duplicate document entries are automatically merged to reduce context redundancy.
The values provided here are common starting points and should be evaluated against the reader’s own samples.