What Data for This Category Looks Like
This category’s work consultation data comes primarily from coal mine safety regulations, job operation manuals, dispatch operation logs, equipment maintenance archives, and emergency response plans. Update frequency varies by document type: safety regulations are revised periodically per industry standards, equipment logs are updated in real time with maintenance operations, and job Q&A documents are supplemented according to team patrol frequency. Document structures include long-text regulations, structured table logs, and short-text job Q&As. Fields cover operation steps, safety thresholds, equipment numbers, and fault codes. Units include industrial standard units such as pressure, weight, and duration.
How These Characteristics Impact Knowledge Base Retrieval and Recall
Long safety regulation texts can lead to overly large default chunk sizes, breaking the complete logic of operation steps. If structured table logs are chunked as plain text, table headers and corresponding business rows will be split, causing matching failures. Real-time updated equipment logs require the knowledge base to support incremental synchronization to avoid data lag. Precise fields such as fault codes and safety thresholds require finer recall granularity to prevent invalid cross-field matching. Most queries in work consultation scenarios focus on specific operations or fault troubleshooting, requiring higher business relevance between recall results and queries.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Most coal mine work consultation queries are precise business questions. This range filters irrelevant small talk and retains valid matching results |
limit | Top 6–10 results | Coal mine scenario queries mostly correspond to single-scenario operations. Too many recall results increase token consumption and interfere with business judgment |
searchMode | vector+keyword | Balances semantic matching and precise matching for structured fields such as fault codes and equipment numbers |
usingReRank | Enabled | There are many professional terms in coal mine business, and reranking further improves the relevance of results to queries |
datasetSearchUsingExtensionQuery | Disabled | Most coal mine work consultation queries have clear business scenarios, and extended queries may introduce irrelevant industry content |
collectionIds | Bound by business module | Only recall knowledge base collections corresponding to coal mine job positions and equipment types, avoiding invalid cross-scenario recall |
Three Common Configuration Mistakes
- Recall results include irrelevant operation steps across job positions, and individual result character counts exceed expectations. Cause: The chunking strategy was not adjusted for coal mine long-text regulations, and default chunk granularity covers multiple unrelated job operation contents.
- Online search triggered after the knowledge base returns no matches returns irrelevant content, or small talk queries are incorrectly recalled. Cause: No
similaritythreshold filter was configured, and no matching success judgment conditions were set based on the business scenario. - Structured table log data splits headers and business rows during recall, making matching results unable to correspond to specific parameters. Cause: The retrieval mode was not adjusted for table data, and default text chunking destroys the structured association of tables.
How to Verify Correct Configuration
- Import a single coal mine job operation manual, run simulated queries, check the character length and business relevance of recall results, adjust chunking and
similarityparameters until they meet expectations. - Input different types of queries (operation consultation, fault troubleshooting, small talk), check the number and matching degree of recall results, verify the configuration effects of
limitandsearchMode. - View knowledge base recall logs, confirm that only
collectionIdsbound to the corresponding business module are included, and no content from irrelevant collections is introduced. - Trigger scenarios where the knowledge base returns no matches, check whether follow-up operations are performed according to preset logic, and verify the configuration of extended queries and filtering rules.
The values provided are common starting points and should be evaluated against respective 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.