What Data for This Category Looks Like
This category serves change management teams comprising R&D, process, quality, production, and procurement personnel at manufacturing enterprises. Data is sourced from PLM, ERP, MES systems, supplier materials, validation test records, and customer special requirement documents. Daily documents include ECN/ECR forms, change assessment reports, drawings and BOM version records, process validation reports, production switching plans, inventory disposal documents, and customer approval documents. Data updates have no fixed cycle, and are triggered each time a process or engineering change occurs. Document fields include change number, version number, effective date, material code, process parameters, validation results, and more. There are no unified fixed units, with formats adjusting based on the corresponding business documents.
Constraints Imposed on Knowledge Base Retrieval and Recall by These Characteristics
Data sources are scattered across multiple business systems. Retrieval requires cross-source integration of different document types, so recall configuration must support multi-dataset association and mixed content retrieval. Data updates have no fixed cycle, so real-time index performance must be maintained, and retrieval must prioritize matching the latest version of change-related content. Document types include structured forms and unstructured drawings, so vector recall logic must adapt to content across different formats. Content associated with changes has strong relevance, so retrieval must accurately match core fields such as change number and version number to avoid recalling irrelevant historical change content. In bidding scenarios, core materials such as customer approval documents and process validation results must be located quickly, so retrieval priority should be tilted toward business key fields.
Configuration Settings
This configuration applies to FastGPT v4.17.0. Specific parameter values can be adjusted based on actual business scenarios.
| Configuration Item | Recommended Setting | Rationale for This Setting |
|---|---|---|
similarity | 0.72–0.80 | Match professional terminology and version information related to changes, and avoid recalling irrelevant historical changes |
limit | 8 entries | Cover multiple types of documents including ECN/ECR, validation reports, and customer files, and adapt to the information density of bidding scenarios |
searchMode | fullTextRecall | Adapt to mixed retrieval requirements for unstructured content such as drawings and BOM, and structured forms |
usingReRank | Enabled | Improve the sorting accuracy of change-related content, and prevent core validation reports from being crowded out by low-relevance results |
datasetSearchUsingExtensionQuery | Enabled | Automatically supplement extended retrieval terms for fields such as change number and version number, and improve recall accuracy |
collectionIds | Specify change-related dataset IDs | Limit the retrieval scope to core change documents such as ECN/ECR and validation reports, and avoid interference from irrelevant data |
Three Common Configuration Mistakes
- Phenomenon: The number of retrieval results is fixed at the preset upper limit. It is not possible to increase recalled content by lowering
similarityor increasing thelimitparameter. Cause: Variable mapping rules for multi-dataset association are not configured correctly, or the system imposes a default limit on the total number of recalled entries per session. - Phenomenon: Retrieval results do not include auxiliary data such as supplier materials and customer special requirements. Cause: The corresponding data source is not added to the associated dataset, or the
datasetSearchUsingExtensionQueryconfiguration is not enabled. - Phenomenon: Empty retrieval results cannot be accurately identified, leading to missing necessary validation documents in bid documents. Cause: Only a basic detector is used to check for empty results, and expired invalid content is not filtered using the change effective date field.
How to Verify a Correct Configuration
- Upload a latest ECN document and a historical change record. After executing retrieval, check whether returned results prioritize displaying the latest version of content.
- Adjust the
similarityparameter to 0.65 and 0.85. Compare changes in retrieval result relevance and quantity to confirm that the configuration takes effect. - Test the multi-dataset associated retrieval process. Verify whether BOM data from the PLM system and inventory disposal documents from the ERP system can be recalled simultaneously.
- Check system logs to confirm that the retrieval request carries the correct
collectionIdsparameter and extended retrieval terms.
The values provided are common starting points and should be evaluated against the reader's own samples.