What data for this category looks like
Daily documents for welded structural component manufacturing include welding procedure qualification records, welding procedure specifications, welding consumable issue records, welder qualification records, weld inspection reports, and repair orders. Data sources include design drawings, material quality certificates, welding equipment parameters, welding consumable warehouses, non-destructive testing systems, and personnel qualification ledgers. Data is updated per weld, per batch, or per shift. Individual data entries include fields such as weld number, welding consumable grade, welder certification number, inspection results, and repair times. Most fields are structured text and numerical values, with units including millimeters, kilograms, hours, and others.
What constraints these characteristics impose on knowledge base retrieval and recall
This category of data has high update frequency and strong per-entry relevance, requiring knowledge base retrieval to support high-frequency incremental updates and associated document recall. Daily documents include both structured and unstructured types, requiring retrieval to adapt to both structured fields such as weld numbers and welding consumable grades, as well as unstructured text such as welding procedure descriptions. Data such as welder qualifications and welding procedure qualification records has rigid constraints, requiring retrieval to accurately match key fields and avoid low-relevance documents interfering with ledger summary. Data sources are scattered across multiple independent systems, requiring retrieval to strictly limit to the specified data source set and prevent cross-category data from mixing into retrieval results.
How to configure the settings
The following configurations are adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Requires accurate matching of rigid constraint fields such as welder qualifications and welding procedure qualification records, to avoid low-similarity irrelevant documents mixing into retrieval results |
limit | Top 8 entries | Associated documents for welded structural components (welding consumables, welders, inspection reports, etc.) have a moderate number; too many will exceed the context window, while too few will miss key associated information |
searchMode | mixedRecall | Requires matching both the text content of welding procedure specifications and structured fields such as weld numbers and welding consumable grades, balancing semantic and exact matching |
usingReRank | Enabled | Requires reordering retrieved results by relevance to avoid low-quality documents interfering with production ledger summary and report generation |
datasetSearchUsingExtensionQuery | Enabled | Requires automatically expanding retrieval keywords based on the user's ledger generation request, to supplement matching structured fields such as weld numbers and welder certification numbers |
collectionIds | Specify the dedicated knowledge base collection ID for welded structural components | Isolate documents from other categories, avoid retrieving irrelevant data from non-welded structural component sources, and ensure precise retrieval scope |
Three common mistakes
- Symptom: The knowledge base retrieval API call returns an unsupported format error, but the corresponding document link can be downloaded normally as a docx file via a browser. Cause: Preprocessing rules for document parsing are not correctly configured, causing the embedded format of the docx file to fail to be correctly recognized by the system.
- Symptom: The number of retrieval results does not match the configured
limitparameter, with either too many or too few retrieved entries. Cause: ThecollectionIdsparameter is not correctly bound, causing the retrieval scope to cover non-target category knowledge base collections. - Symptom: Unable to configure a workflow node where users can choose whether to skip knowledge base retrieval. Cause: The workflow's variable configuration function is not enabled, and the retrieval switch is not bound as a selectable parameter for users.
How to confirm the configuration is complete
- Upload a test welding consumable issue record docx document, initiate a retrieval request, and verify that the document can be retrieved.
- Check the
similarityscore of the retrieval results, confirm that the score falls within the configured0.75–0.85range. - Check the number of retrieval results, confirm that it matches the configured
limitparameter. - Trigger a production ledger generation request, verify that relevant documents such as welding procedure qualification records and welder qualification records can be automatically associated and retrieved.
The values provided are common starting points and should be measured against the reader's own samples.