Data in this category
Data sources include monitoring alerts, log platforms, APM tools, and user feedback. Daily documents cover incident tickets, event timelines, emergency response plans, and fault notifications. Data updates occur in real time or sync after each incident. Most documents use structured fields, including incident ID, impact scope, occurrence time, and troubleshooting operation records. Only raw records and disposal content from the full incident lifecycle are retained. No custom statistical fields are added.
Constraints for knowledge base retrieval and recall
Since data sources include real-time monitoring and user feedback, retrieval must support access to dynamically updated datasets. Outdated disposal plans must not be returned in results. Documents contain specific fields such as incident ID and impact scope. Retrieval must match these fields to accurately locate relevant incident tickets and emergency plans. The real-time update feature requires recall results to sync with current incident status. Static cached data cannot be used. Core time-consuming workflows in this category depend on accurate disposal references. Retrieval must prioritize content strongly related to the current fault’s impact scope and troubleshooting operations. Redundant results must not interfere with decision-making.
Configuration settings
These configurations are based on FastGPT v4.17.0, and are tailored to knowledge base retrieval needs for incident response and troubleshooting scenarios.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the semantic relevance requirements for fault disposal documents in this category. Prevents recall of irrelevant historical incident records, and ensures accuracy of disposal references |
limit | Top 3–5 entries | This category requires quick access to core disposal references. Too many results increases decision-making burden |
searchMode | hybrid | Combines keyword and semantic retrieval. Adapts to content in incident tickets that has both clear identifiers and semantic associations |
usingReRank | Enabled | Performs secondary sorting on initial recall results. Prioritizes response plans that better match the current fault’s impact scope |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands relevant fault disposal keywords based on user queries. Covers vague questions from operations staff |
collectionIds | Exclusive collection ID of the target fault knowledge base | Isolates business data, and prevents retrieval of documents from non-operations scenarios |
Three common configuration errors
- Phenomenon: After configuring the specified embedding model and knowledge base parameters, retrieval returns a "no matching results" error. Cause: The
collectionIdsof the target fault knowledge base were not added to the corresponding configuration item, or the embedding model configuration was not linked to the exclusive fault knowledge base dataset. This means the retrieval scope does not cover business-specific data. - Phenomenon: The first reference content from the knowledge base is correct, but it is not displayed in the response body. Cause:
datasetSearchUsingExtensionQuerywas not enabled. The reference content cannot be linked to the query’s semantics, so the generated response does not integrate the reference content. - Phenomenon: After initiating a conversation via the API, the generated reply does not include knowledge base document content. Cause: The associated
collectionIdswere not specified in the request parameters, orsearchModewas not set to hybrid. This means relevant knowledge base content cannot be recalled.
How to verify correct configuration
- Log in to the FastGPT admin backend, and confirm that the
collectionIdsin the knowledge base configuration match the collection identifier of the target fault knowledge base. - Submit a test query related to the current operations scenario. Check that the number of retrieval results matches the setting specified by the
limitconfiguration. - Enable retrieval debug logs, and confirm that parameters such as the
similaritythreshold andusingReRankswitch are loaded correctly. - Call the official API interface to initiate a conversation, and verify that the returned results include knowledge base document content within the configured scope.
The values listed are common starting points, and should be evaluated against the specific deployment’s own datasets.
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.