Data Profile for This Category
Data for this category primarily originates from customer ITSM systems, monitoring platforms, asset inventories, and on-duty records. Four document types are generated regularly: daily on-site reports, service desk tickets, inspection reports, and SLA monthly reports. Most data is updated daily; some SLA monthly reports are updated monthly. Documents include fields such as on-site staff information, ticket numbers, inspection items, and SLA compliance rates. There are no unified fixed units, and some fields must match the format of the customer’s existing systems.
Constraints on Knowledge Base Retrieval and Recall
Scattered data sources and inconsistent formats increase preprocessing complexity, as knowledge base imports must adapt to multiple source system structures. Differences in update frequencies require prioritizing the most recent daily tickets and on-duty records during retrieval, to avoid calling outdated monthly SLA data. Diverse document fields require targeted retrieval rules configured for scenarios including on-site staff, tickets, and inspections, to prevent irrelevant fields from interfering with recall results. Additionally, the contracted delivery nature of on-site services requires retrieval results to align with service agreement response boundaries, avoiding content recalled outside the service scope.
Configuration Recommendations
These recommendations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches text similarity for on-site tickets and inspection reports, balances recall precision and coverage |
limit | Top 8 entries | Controls the number of single-recall results, matches the information load that on-site engineers can handle during on-site work |
searchMode | Semantic search | Adapts to unstructured documents from multiple sources, better suited for colloquial content in ticket descriptions |
usingReRank | Enabled | Performs secondary sorting on recall results, filters redundant low-relevance content from on-site daily reports |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval terms based on user questions, covers synonymous expressions of professional terms in inspection reports |
collectionIds | Dataset IDs bound to service contracts | Limits retrieval scope, ensures only on-site service documents for the corresponding customer are called |
Three Common Misconfigurations
- Phenomenon: Incomplete knowledge base recall coverage, with user questions related to on-site tickets not being matched. Cause:
datasetSearchUsingExtensionQueryis not enabled, or extended retrieval terms are not configured for the ticket scenario, so synonymous expressions cannot be identified. - Phenomenon: No matching results when specifying a knowledge base tag in the
api/v1/chat/completionsAPI call. Cause: ThecollectionIdsparameter is not correctly bound, or the tag does not match the dataset configuration. - Phenomenon: Unable to enable the image understanding function when creating a general knowledge base. Cause: The local deployment version lacks corresponding model dependencies. The image understanding model package must be deployed additionally.
How to Confirm Successful Configuration
- Submit typical on-site ticket questions to the retrieval interface, and verify that the similarity scores of returned results fall within the preset range.
- Check that the number of single-recall results matches the configuration of the
limitparameter. - Confirm that after specifying
collectionIds, only content from datasets bound to the corresponding service contract is returned. - After enabling
usingReRank, confirm that the sorting priority of recall results meets the requirements of the business scenario.
The values provided here are common starting points and should be measured against samples from the actual deployment.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.