What the Data for This Category Looks Like
Core data for this category comes from ERP orders and inventory, bills of materials (BOMs), process routes, equipment capacity, staff shifts, work-in-progress, and supplier deliveries. Daily processed documents include sales orders, master production schedules, material requirements plans, weekly production schedules, work orders, material shortage lists, and delivery change orders. These documents typically contain fields such as order numbers, material codes, capacity values, shift durations, and process numbers, with units including pieces, hours, days, and more. Data update frequency is usually daily, weekly, or synchronized when orders or exceptions occur. The following content is written based on the configuration logic of FastGPT v4.17.0.
Constraints Imposed by These Characteristics on Knowledge Base Retrieval and Recall
Multi-source heterogeneous data sources require the retrieval link to support connections to different types of business system documents, to avoid missing key scheduling reference information. The high-frequency update rhythm triggered by exceptions requires the knowledge base retrieval to support incremental synchronization and partial updates, without needing to fully rebuild the vector database. Diverse document structures with a large number of structured fields require retrieval to balance semantic matching and precise field matching, to ensure recalled content is strongly relevant to current scheduling tasks. Long-text production schedules and work order documents also require chunking and vector generation links to adapt to long-content processing logic, to avoid semantic fragmentation.
How to Configure Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Production planning and scheduling documents have numerous fields and are highly interconnected. Low-relevance redundant results must be filtered to ensure the accuracy of recalled content |
limit | Top 6–10 entries | Scheduling tasks require reference data that balances multiple constraints such as equipment, materials, and personnel. An appropriate number of recalled entries can cover key information without increasing inference load |
searchMode | mixedRecall | Matches both semantic relevance and structured fields such as order numbers and work order IDs, adapting to the mixed retrieval needs of production planning documents |
usingReRank | Enabled | Re-ranks initial recalled results by relevance, prioritizing work order and capacity data most relevant to the current scheduling scenario, improving the efficiency of subsequent AI responses |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands the user's scheduling requirements into queries associated with work orders, equipment status, and staff shifts, covering more relevant business documents |
collectionIds | Specify dedicated collection IDs for production planning and shop scheduling | Limits the retrieval scope to only business documents of this category, avoiding recalling irrelevant equipment maintenance content and improving retrieval accuracy |
Three Common Mistakes
- Symptom: The number of retrieval results is far lower than the set
limitvalue. Cause:collectionIdsis not configured correctly, and the corresponding production planning knowledge base collection is not specified, resulting in only partial data sources being recalled. - Symptom: AI conversation output takes too long. Cause:
usingReRankis not enabled and thelimitvalue is set too large, which loads a large number of low-relevance scheduling documents and increases context splicing and inference time. - Symptom: An error indicating exceeding model length limit is reported during vector generation. Cause: Chunking parameters are not adjusted for long production planning documents, and the general chunking length is used directly, resulting in a single chunk exceeding the vector model's length limit.
How to Confirm the Configuration Is Correct
- Manually upload a test sales order document, execute a scheduling query associated with equipment inspection and maintenance, and check whether the retrieval results include the key content of this document.
- View the knowledge base update log to confirm that newly added documents such as delivery change orders and material shortage lists are automatically synchronized at the preset frequency, with no omissions or delays.
- Adjust the
similarityparameter to verify whether the relevance of retrieval results meets the needs of the current scheduling scenario, and there is no need to use a fixed uniform value. - Test the partial update function for a single document. After modifying part of the content, confirm that only the modified fragment needs to be uploaded to complete the knowledge base synchronization, without full replacement.
- After enabling
datasetSearchUsingExtensionQuery, execute a query for temporary order insertion requests, and confirm that associated equipment capacity and staff shift documents are recalled.
The values provided are common starting points and should be measured against the reader's own samples.