Data Profile for This Category
The category primarily covers bread, pastry, biscuit, instant noodle, and frozen pastry manufacturing plants. Daily operation and production documents include formula and feeding lists, production records, baking or steaming curves, metal detection records, cleaning and disinfection records, and factory inspection reports. Data sources cover raw and auxiliary material qualification certificates, ERP/MES systems, weighing systems, oven or frying line equipment, laboratory test results, and food traceability ledgers. Data update frequency follows production batches, team shifts, and daily cycles. Core time-consuming steps focus on formula and allergen material review, proofing/baking/frying window control, and foreign matter and packaging shelf life verification.
Constraints on Knowledge Base Retrieval and Recall
Large numbers of formulas and frequent production line changeovers require retrieval and recall to quickly match the latest single-batch production parameters and cleaning records, preventing expired or mismatched documents from being returned. High-priority requirements for allergens and cleaning switching require retrieval results to accurately distinguish exclusive documents for different production lines, reducing cross-line information interference. Mixed data from multiple sources (structured parameters and unstructured text) require retrieval modes to balance semantic understanding and keyword matching, covering content in different formats such as curve values and feeding lists. The high-frequency update feature requires the knowledge base index to quickly synchronize the latest data from each shift and batch, avoiding retrieval results lagging behind actual production status.
Configuration Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.78 | For high-precision scenarios such as matching allergen labels and equipment parameters, calibrated based on actual testing |
limit | Top 8 entries | Covers retrieval needs for multi-batch production records and multi-line cleaning records, avoiding missing key information |
searchMode | mixedRecall | Adapts to hybrid retrieval scenarios for formula text and baking curve values |
usingReRank | Enabled | Re-sorts retrieval results to optimize recall priority for allergen materials and foreign matter control records |
datasetSearchUsingExtensionQuery | Enabled | Expands question boundaries to match content such as proofing windows and feeding parameters in production records |
collectionIds | Configured by production line group | Differentiates knowledge base collections for different baking and frying production lines, enabling precise recall of inspection and maintenance documents for the corresponding line |
Three Common Misconfigurations
- Phenomenon: The knowledge base export file is only hundreds of KB, with incomplete content. Cause: Did not select full dataset export, only exported documents from partial grouped collections.
- Phenomenon: Document content with local image links cannot be displayed in retrieval results. Cause: Did not upload local images to a publicly accessible online image hosting service and replace the original links; the knowledge base only supports image resources that can be loaded online.
- Phenomenon: Insufficient retrieval accuracy, with incorrect matches for allergen materials. Cause: The
similaritythreshold is set too low, resulting in low-relevance results being recalled, orcollectionIdsare not configured by production line group, leading to cross-line document interference.
How to Confirm Correct Configuration
- Upload a test baking curve document, initiate a retrieval containing "baking temperature", confirm that the key parameters of this document are included in the results.
- Trigger a full knowledge base retraining, wait for training to complete, check the dataset status, confirm that all documents have been successfully indexed.
- Initiate a retrieval containing a known allergen material name, check whether the sorting of results conforms to business priorities.
- Export the dataset for a specified group, check the content of the exported file against the number of original documents, confirm that the export range meets expectations.
The values provided are common starting points and should be evaluated against the samples relevant to the specific deployment.