What the data for this category looks like
Daily documents for this category include backup policies, recovery drill reports, disaster recovery switchover plans, and data verification records. Data sources cover backup systems, database logs, object storage, and disaster recovery platforms. Document fields include Recovery Point Objective (RPO), Recovery Time Objective (RTO), drill time, verification results, and others. Some fields use units such as hours and count. Data update frequency is divided into daily, monthly, and quarterly categories based on backup cycles. Backup system logs are updated daily. Quarterly disaster recovery drill reports are generated quarterly. The overall data structure covers information related to full-link recoverability.
What constraints these characteristics impose on the reranking and hybrid retrieval link
Data sources for this category are scattered and have diverse structures. Hybrid retrieval must integrate logs and documents from multiple systems, so both semantic and keyword features must be matched. Update frequencies vary significantly across different documents, so expired old backup policies or incomplete drill reports must be avoided in retrieval. Special fields such as Recovery Point Objective (RPO) and Recovery Time Objective (RTO) need to be core dimensions for relevance matching. The reranking link must prioritize sorting drill reports and verification records that match the current consultation scenario. Cross-system associated data requires retrieval results to cover the full backup, storage, and disaster recovery chain, and cannot be limited to content from a single data source.
Configuration guidelines
This page's configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
usingReRank | Enabled | Documents in this category have high professionality and scattered sources. Reranking filters low-relevance redundant content and improves content accuracy for pre-sales Q&A |
similarity | 0.75–0.85 | Operational documents have high terminology density. A threshold that is too low will introduce irrelevant general documents, while a threshold that is too high will miss key recovery drill reports |
limit | Top 8–12 entries | The context window for pre-sales Q&A is limited. Too many retrieval results will exceed the model's processing range. The top 8–12 entries cover core backup and disaster recovery-related documents |
searchMode | hybrid | Data for this category comes from multiple systems. Hybrid retrieval combines semantic matching and keyword matching, covering both keywords from backup logs and semantic content from drill reports |
Three common configuration errors
- Symptom: After setting
usingReRankto Enabled, the order of knowledge base retrieval results does not change. Cause: The corresponding reranking model is not bound in the FastGPT backend, or the reranking model's call parameters are not correctly configured. - Symptom: The reranking model configuration option cannot be selected when creating an application. Cause: The current FastGPT version does not have the reranking function enabled, or the reranking model list is not correctly declared in the configuration file.
- Symptom: The number of returned retrieval results does not match the configured
limit. Cause: Multi-data source recall is enabled during hybrid retrieval, and the recall results are not truncated by the reranking link, causing the final returned number of entries to exceed the set value.
How to confirm configurations are correctly applied
- Enter the knowledge base management page of the FastGPT application, check the enabled status of the reranking function, and confirm that
usingReRankis enabled. - Enter a query related to backup and recovery, such as "How to verify disaster recovery recovery consistency", and check whether retrieval results prioritize displaying documents such as recovery drill reports and data verification records.
- Adjust the
similaritythreshold, observe changes in the relevance of retrieval results, and verify whether the threshold configuration takes effect. - View the application's running logs, confirm that the reranking model's call requests have been sent normally, and no error messages are returned.
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.