What this category of data looks like
Data for this category originates from PLM, ERP, MES, supplier materials, validation tests, and customer-specific requirements. Daily documents include ECN/ECR, change assessment reports, drawings and BOM version files, process validation reports, switching plans, inventory disposal forms, and customer approval documents. Updates occur irregularly, triggered by each change process. Document structures include fixed fields such as change number, version number, effective date, associated material/process node, and approval status. Some documents include binary drawing attachments.
What constraints these characteristics impose on knowledge base retrieval and recall
Multi-source heterogeneous data sources require retrieval to support cross-system associated matching. This prevents missing cross-departmental change-related information. Irregular update rhythms require configuring incremental synchronization mechanisms. This ensures the knowledge base only contains the latest versions of change documents. Documents include fixed fields such as change number and version number. Retrieval must prioritize exact field matching to avoid recalling irrelevant old version content. Associated upstream and downstream impacts require retrieval to cover associated materials and process nodes. This ensures the completeness of quality traceability.
How to configure the settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Change-related documents have high semantic similarity requirements, to avoid recalling irrelevant old version change content |
limit | Top 6 entries | Change documents have many associated items. Too many single recalls will exceed the context window and affect answer coherence |
searchMode | mixedRecall | Change documents have both semantic associations and clear keywords such as change number and version number. Hybrid mode can improve accuracy |
usingReRank | Enabled | Filter recall results that are semantically similar but do not match versions, to ensure only content related to the target change is returned |
datasetSearchUsingExtensionQuery | Enabled | Automatically expand upstream and downstream material and process keywords related to changes, to cover the retrieval scope of chain impacts |
collectionIds | Specify the knowledge base collection IDs corresponding to PLM and MES | Limit the retrieval scope, avoid mixing non-change-related documents, and resolve mandatory field verification errors |
Three common configuration mistakes
- Phenomenon: A
Validation failed: name: Path 'name' is requirederror appears when viewing knowledge base citations in chat responses. Cause: ThecollectionIdsparameter is not configured, the retrieval target knowledge base collection is not limited, and the system cannot match the mandatory collection name field. - Phenomenon: Unexpected knowledge base search input and response citation content appears in responses. Cause: The number of recall results is not controlled via the
limitparameter, or unnecessary retrieval log output is not disabled. - Phenomenon: Recall results include old version documents unrelated to the current change. Cause: The
similaritythreshold is set too low, or the keyword matching mode ofsearchModeis not enabled, and content that does not match versions is not filtered.
How to confirm the configuration is complete
- Enter the knowledge base management interface of FastGPT v4.17.0, verify that the
collectionIdsparameter has been filled with the knowledge base collection IDs corresponding to PLM and MES. - Initiate a test query containing a change number, check whether the
similarityscore of the returned results falls within the preset 0.75–0.85 range. - View the citation list of the response, confirm that only documents of the target category such as ECN/ECR and process validation reports are included, with no irrelevant content.
- Trigger an incremental synchronization of change documents, check whether the knowledge base can normally update the latest change version content.
The values provided are common starting points and should be evaluated against the reader's own samples.