What the data for this category looks like
This category of data primarily comes from customers’ legacy systems, CSV or Excel templates, database exports, and product import logs. Daily associated documents include data templates, migration plans, field mapping tables, and migration verification reports. Data update cycles are per individual customer or irregular. Core content revolves around business data identifiers, field mapping relationships, data validation rules, and import result statuses. There are no unified fixed units; corresponding fields and units must be matched based on the customer’s specific business scenarios.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The data sources for this category are scattered and have diverse formats, including both structured field mapping tables and CSV templates, as well as unstructured migration plans and verification reports. This requires the retrieval link to support recall of multi-format data. Data updates are per customer and have no fixed cycle, so support for on-demand synchronization of the latest migration configurations and log documents is needed. Core business documents focus on field mapping and data validation, requiring retrieval to accurately match professional terms such as field names and mapping rules, and avoid recalling irrelevant content. At the same time, error messages in import logs must be quickly linked to corresponding solution documents, so recall accuracy needs to be optimized to meet the timeliness requirements of technical support scenarios.
How to set configurations
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches professional terms such as field mappings and migration rules, balancing recall coverage and accuracy |
limit | Top 6 entries | Covers common migration issue scenarios, avoiding returning excessive redundant content |
searchMode | Semantic search | Adapts to mixed retrieval needs for semi-structured migration plans and structured field mapping tables |
usingReRank | Enabled, re-rank top 3 entries | Improves the sorting priority of core error reports and field mapping-related content |
datasetSearchUsingExtensionQuery | Enabled | Supplements associated extended content such as migration logs and verification reports, covering extended questions in pre-sales Q&A |
collectionIds | Bind a dedicated collection for migration-related documents | Isolates knowledge base content from other categories to avoid recalling irrelevant documents |
Three common misconfigurations
- A "Request failed with status code 400" error occurs because the
collectionIdsdedicated migration document collection is not configured, and cross-source data format verification exceptions are triggered during retrieval. - Knowledge base retrieval fails to match migration-related questions associated with titles, because
datasetSearchUsingExtensionQueryis not enabled, and document titles are not included in the retrieval scope. - An exception of automatic Bearer token supplementation occurs when calling the API knowledge base, because the authorization parameter logic in the retrieval configuration is not adjusted, causing the request header format to not meet the requirements of the target system.
How to verify correct configuration
- Manually input keywords related to field mapping, and verify that returned results include dedicated documents for this category.
- Simulate an import scenario to trigger error keywords, and confirm that corresponding migration verification or troubleshooting documents are recalled.
- Confirm that the
collectionIdsconfiguration only binds migration-related knowledge base collections, eliminating interference from documents of other categories. - Verify that after
usingReRankis enabled, content related to core errors is sorted ahead of redundant content.
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.