What the data for this category looks like
The data sources for this category are change management systems, CI pipelines, monitoring data, and business calendars. Daily generated documents include change requests, risk assessments, launch window plans, and rollback plans. These documents contain fields such as change ID, risk level, execution time window, rollback trigger conditions, and more. The data update frequency is weekly or after each completed change, and documents related to a single change are updated synchronously before and after the change is executed.
What constraints these characteristics impose on the reranking and hybrid retrieval link
Change management data includes both structured metadata and unstructured document content. This requires hybrid retrieval to balance semantic matching and precise field matching, to avoid recalling irrelevant operation and maintenance tickets. The data update frequency is weekly or after a single change, which requires the retrieval index update cycle to match the data update rhythm, to avoid retrieving expired change window or risk assessment content. Documents contain priority fields such as risk level and execution time window. The reranking link needs to prioritize content with high priority related to the current on-call scenario, while adapting to format differences across data sources to ensure the availability of retrieval results.
How to set the configurations
These configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
usingReRank | Enabled | Change management data includes priority fields such as risk level and execution time window. Enabling reranking allows prioritization of high-priority change-related content |
similarity | 0.75–0.85 | Semantic similarity requirements for change-related tickets and documents are high. This range can filter out irrelevant non-change content |
limit | Top 8 entries | Control the number of results returned per retrieval within a reasonable range, adapting to the quick review needs of on-call operation and maintenance personnel |
searchMode | Hybrid Retrieval | Change management data includes both structured metadata and unstructured documents. Hybrid retrieval balances semantic matching and precise field matching |
Three common mistakes
- Phenomenon: Reranking result fields display as false. Cause: The reranking model is not bound to the current knowledge base, or the configured reranking model does not support the index format of the current knowledge base.
- Phenomenon: The original configuration is automatically overwritten after adding a reranking model with the same name. Cause: The platform only allows one configuration for reranking models with the same name. Adding a configuration with the same name will replace the old configuration.
- Phenomenon: Calling the reranking model returns empty content. Cause: The API access key for the model is not configured correctly, or reasonable call timeout parameters are not set.
How to confirm the configuration is complete
- Enter a ticket question containing "change window" or "risk assessment" on the knowledge base retrieval test page, and check whether the returned results include relevant change documents.
- Go to the reranking model configuration page, confirm that the currently used model name is not overwritten by other configurations, and the model status shows normal.
- Check the retrieval log, confirm that the
usingReRankparameter has been enabled, and the returned reranking result fields are not empty. - Enter a specific change ID as the retrieval keyword, and check whether the retrieval results prioritize matching change-related documents corresponding to that ID.
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 2026-10-19.