What this category of data looks like
This category of data primarily comes from customer historical systems, CSV or Excel templates, database exports, and product import logs. It is maintained daily by implementation consultants, data implementation engineers, and customer IT liaisons. Daily produced documents include data templates, migration plans, field mapping tables, and migration verification reports. The data update rhythm follows per-order or irregular schedules. Each individual data set includes field mapping rules, batch import parameters, dirty data cleaning standards, and other content. There is no fixed unified unit, and adjustments are made based on the customer's business scenario.
What constraints these characteristics impose on knowledge base retrieval and recall
Dispersed data sources and diverse formats require knowledge base retrieval to support field matching logic for multi-format data sources, and avoid cross-system field misalignment. Irregular update cycles for individual data sets require flexible switching between incremental recall and full refresh. Daily documents have fixed structures but complex field logic, requiring precise recall of associated documents such as field mapping tables and migration plans, to prevent retrieval results from mismatching the current ticket scenario. The volume of batch import log data fluctuates with ticket scale, requiring retrieval response constraints to be adapted for different concurrency levels.
How to set configurations
The following configurations are formulated based on the knowledge base function of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Matches professional documents such as field mapping tables and migration plans. Prevents retrieval of irrelevant content when set too low, and avoids missing valid information when set too high. This interval is calibrated based on actual testing. |
limit | Top 6 entries | The number of documents associated with a single ticket is usually limited. Controlling the number of returned entries avoids context overload. |
searchMode | embedding | Adapts to the semantic matching requirements of structured documents such as field mapping tables, and prioritizes recall of migration-related documents with semantic relevance. |
usingReRank | Enabled | For long-text documents such as batch import logs and field mapping tables, re-ranking optimizes the sorting priority of highly relevant results. |
datasetSearchUsingExtensionQuery | Enabled | Supplementally matches extended keywords such as customer system names and field names mentioned in user tickets, improving retrieval accuracy. |
collectionIds | Specify the dedicated collection for customer data import and migration | Isolates business scenarios, preventing retrieval results from being disturbed by documents from other categories. |
Three common misconfigurations
- Issue: Retrieval results return documents not related to customer data import and migration. Cause: The
collectionIdsparameter is not specified, so the retrieval scope covers the entire platform knowledge base. - Issue: Associated documents such as uploaded field mapping tables and migration plans are split into multiple independent entries, and cannot be matched as a whole against user tickets. Cause: Associated grouping parameters are not configured correctly, or the chunking logic does not retain the structural association of the document.
- Issue: After calling the knowledge base interface to create a training order, the retrieved recall results do not include the expected import log data. Cause: The
trainingTypevalue does not match the type requirements of plain-text business documents, or thesimilaritythreshold is set too high, causing valid documents to not be recalled.
How to confirm configurations are properly set
- Manually enter a test question related to field mapping, and verify that returned results only include documents related to customer data import and migration.
- Upload a set of test documents with chapter titles and links, and verify that they are recalled as a whole without being split.
- Simulate concurrent submission of retrieval requests for multiple tickets, confirm that there is no significant delay in retrieval response, and that concurrency support meets business expectations.
- Call the knowledge base interface to create a training order, check that the returned order status is normal, and that the parameter configurations of associated documents meet requirements.
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.