What the data for this category looks like
This category of data primarily comes from enterprise ERP systems, process documents, equipment ledgers, and scheduling systems. Daily generated documents include sales orders, master production schedules, material requirements plans, weekly and daily production scheduling tables, work orders, material shortage lists, and delivery change notices. Data update frequency varies by business scenario: equipment capacity and staff shift data is synced daily, master production plans are updated weekly, and relevant documents are updated immediately when exceptions such as order changes, material shortages, or equipment failures occur. Most documents contain structured fields such as order numbers, material codes, capacity values, and delivery times. Some process documents also include process parameter descriptions and drawing attachments.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
The data characteristics of this category create clear constraints for retrieval and recall. Decentralized storage across multiple data sources requires the retrieval pipeline to support unified recall across ERP systems, process documents, and scheduling systems to avoid data silos. Differentiated update frequencies require configuration support for incremental sync and scheduled full updates to ensure recalled scheduling data matches current business status. Structured business fields require retrieval to match field semantics. Relying solely on keyword matching cannot accurately target business-related content. Cross-process, cross-material retrieval requirements mean recall results must cover multi-dimensional information such as resource constraints, delivery times, and material inventory to meet the real-time decision-making needs of shop scheduling.
How to set the configuration
This configuration is based on FastGPT v4.17.0, adapted to the specific needs of production planning and shop scheduling scenarios.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches semantic relevance of structured business fields, avoids recalling irrelevant general documents |
limit | Top 10 entries | Covers multiple document types including work orders, scheduling tables, and bill of materials, avoids excessive results exceeding the context window |
usingReRank | Enabled | Re-ranks recall results by business relevance, prioritizes accurate content containing order numbers and capacity values |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands search terms to cover business-specific terminology such as material codes and process names, improving recall coverage |
searchMode | mixedRecall | Combines keyword matching and semantic retrieval, adapts to mixed content of structured documents and process descriptions |
collectionIds | Bind to dedicated knowledge base collections for production planning and shop scheduling | Avoids recalling documents unrelated to this category, narrows retrieval scope to improve accuracy |
Three Common Mistakes
- Retrieval results do not include process drawings or image content within documents. Cause: The knowledge base's image parsing and embedding configuration is not enabled, only text content is extracted.
- Unable to enable online search functionality after local deployment. Cause: No proxy or interface permissions for online search are configured, preventing retrieval requests from being sent normally.
- The number of retrieval return results does not match the set
limitparameter. Cause: Redundant recall filtering rules are not turned off, or the number of valid documents matching in the knowledge base is insufficient.
How to Confirm the Configuration is Correct
- Upload test documents containing order numbers and material codes, perform test queries, and verify that the returned results include the aforementioned specific business fields.
- View retrieval logs to confirm that the number of recall entries triggered by the
similaritythreshold falls within the preset range, and that the re-ranked results have the correct priority. - Check the bound collection IDs of the knowledge base to confirm that only documents related to production planning and shop scheduling are included, with no unrelated content mixed in.
- Upload process documents containing images, perform queries that include image descriptions, and verify that the results return images or text descriptions corresponding to the images.
The values provided are common starting points and should be measured against the reader's own samples.