This document is written for FastGPT v4.17.0. The content below applies to this version and compatible versions.
What the data for this category looks like
The data for this category primarily comes from scheduling systems, fault work orders, monitoring alerts, and organizational contact lists. Update frequency is daily or weekly. Daily documents include shift schedules, handover records, escalation contact lists, and shift daily reports. Fields cover shift time periods, pending unresolved items, escalation contact names and contact information, fault response paths, and more. Data structure adjusts based on team size. Small teams can combine all documents into a single shift handover form. Large teams can split out separate escalation contact lists and shift daily reports.
What constraints these characteristics impose on knowledge base retrieval and recall
Since data updates daily or weekly, knowledge base retrieval must support filtering by update time. This ensures the most recent shift schedules and handover records are retrieved. Different types of business documents correspond to separate knowledge base collections. Retrieval must target specific collections to avoid mixing up shift schedules and escalation contact information from different time periods. Structured content in fields such as shift time periods and unresolved items requires retrieval to support exact field matching, combined with semantic recall to supplement matching of unstructured content. Unresolved items are linked to real-time fault work orders. Retrieval must link to the latest work order data. This ensures returned handover content includes current pending items. Differences in document structure based on team size require retrieval configurations to support flexible adjustment of collection scope. This adapts to different teams’ document organization methods.
How to configure the settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Balances semantic matching accuracy and recall coverage, avoids incorrectly retrieving shift information from non-matching time periods |
limit | Top 8–12 entries | Single retrieval returns content that covers core items for a single round of shift handover, while controlling context length |
searchMode | Vector recall + keyword hybrid | Meets matching requirements for both structured fields such as shift time periods and unstructured content such as unresolved item descriptions |
usingReRank | Enabled | Sorts recall results by relevance, ensures the most relevant shift schedules and escalation contact information is returned first |
collectionIds | Specify collection IDs corresponding to shift schedules, handover records, and escalation contact lists | Targets retrieval of target business documents, avoids mixing in unrelated fault work orders or monitoring alert data |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval keywords, supplements matching of synonymous expressions for structured fields such as shift time periods and contact names, improves recall accuracy |
Three common configuration mistakes
- Symptom: Retrieval results only return a small amount of target data, and cannot match specified shift personnel or handover items. Cause: The corresponding business collection is not specified via
collectionIds, leading the retrieval scope to cover unrelated historical data. No field matching rules are configured to filter non-target content. - Symptom: Retrieval results include expired shift schedules, and current shift personnel information cannot be obtained. Cause: No retrieval condition for filtering by update time is set. The knowledge base has not synced the latest daily scheduling data, leading to retrieval of outdated documents.
- Symptom: Returned handover records do not include complete unresolved item content. Cause:
datasetSearchUsingExtensionQueryis not enabled, so synonymous expressions for unresolved items cannot be matched. Thelimitvalue is set too low, truncating core handover content.
How to verify correct configuration
- Manually input keywords for a specified shift time period. Retrieval results are checked to confirm they only include documents from the target collection. The
collectionIdsconfiguration is verified to take effect. - A test expired shift schedule is uploaded. Retrieval is initiated, and results are checked to confirm they do not include the expired document. The update time filtering configuration is verified to take effect.
- A retrieval term that includes an unresolved item description is input. Returned results are checked to confirm they include complete pending handover content. The reranking and query expansion configurations are verified to take effect.
- The
similarityvalue range is adjusted. Changes in the matching accuracy of retrieval results are observed. Configuration parameters are confirmed to meet business requirements.
The values provided are common starting points, and should be evaluated against samples relevant to the specific use case.
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.