What data looks like for this category
The data for this category comes from ERP orders and inventory, BOMs, process routes, equipment capacity, staff shifts, work-in-progress, and supplier deliveries. Daily documents include sales orders, master production schedules, material requirement plans, weekly and daily scheduling tables, work orders, material shortage lists, and delivery change notices. Data fields include order numbers, process step numbers, material codes, capacity units, delivery dates, and more. Update frequencies are daily, weekly, per order, or when an exception occurs.
What constraints do these characteristics impose on knowledge base retrieval and recall
Data sources are scattered and cover multiple types of business documents, requiring precise limitation of the retrieval scope to avoid mixing non-production data. The update rhythms of different documents vary greatly, so support for business-triggered incremental updates is needed. Structured scheduling tables and unstructured process text coexist, requiring retrieval modes to balance semantic matching and precise keyword matching. Retrieval for specific business fields such as work order ID, delivery date, and capacity value must adapt to field-level precise matching logic to avoid recalling irrelevant content.
How to set configurations
The following configuration items are based on the knowledge base retrieval logic of FastGPT v4.17.0, adapted to the business scenario of production planning and shop scheduling.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for production plans and exception reports, filters low-correlation non-scheduling data |
limit | Top 10 entries | Adapts to the context length limit of shop scheduling scenarios, avoids excessive redundant information interfering with scheduling decisions |
searchMode | mixedRecall | Balances keyword matching for structured scheduling tables and semantic matching for unstructured process documents |
usingReRank | Enabled | Performs secondary ranking on initial recall results to improve the relevance of content related to exception reports |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval terms to cover precise matching for business fields such as work order numbers, delivery dates, and material codes |
collectionIds | Bind IDs of production plan libraries and process document libraries | Limits the retrieval scope, only executes retrieval in specified production-related knowledge bases |
Three common misconfigurations
- Phenomenon: When calling knowledge base retrieval, only the content of the first knowledge base is returned, or automatic switching to the second knowledge base fails. Cause: The retrieval priority logic for multiple knowledge bases is not configured, and the process of first retrieving the production plan library and then the process library is not set according to business needs.
- Phenomenon: The interface prompts "Dataset does not exist" or "Insufficient retrieval permissions". Cause: The ID of the corresponding knowledge base is not correctly filled in
collectionIds, or the bound knowledge base does not have retrieval permissions enabled. - Phenomenon: The retrieval results contain a large number of irrelevant equipment maintenance documents. Cause: The retrieval scope is not limited via
collectionIds, or thesimilaritythreshold is set below 0.7, recalling irrelevant non-scheduling data.
How to confirm the configuration is complete
- Input the test question "2024 October Week 3 machining shop scheduling table", check whether the returned results include the content of the corresponding weekly and daily scheduling tables.
- Input the question "M12 workpiece milling process parameters" which falls outside the scope of the first knowledge base, check whether the second knowledge base is automatically retrieved and the corresponding process document content is returned.
- View the retrieval log to confirm that the
similaritythreshold andlimitnumber of entries meet the configuration requirements, and no out-of-scope recall results exist. - Test the exception scenario by inputting "Work order W00123 material shortage", check whether the corresponding material shortage list and scheduling adjustment plan are recalled.
The values provided here are common starting points and should be measured against the reader's own samples.