What Data for This Category Looks Like
Core data for cosmetics manufacturing enterprises comes from registered or filed formulas, raw material safety documents, weighing and emulsification equipment logs, ERP/MES systems, laboratory test records, and packaging material version documents. Documents generated during daily production include formula work orders, weighing and feeding records, emulsification or filling records, cleaning and sanitization records, semi-finished and finished product test reports, and label review documents. Data updates are triggered by each production batch or each line change. Documents include fields such as formula components, restricted substance content, production batch numbers, and equipment operating parameters. Units include grams, degrees Celsius, batch numbers and other production-specific identifiers.
Constraints Imposed on Knowledge Base Retrieval and Recall
Cross-review of technical documentation for cosmetics manufacturing requires formulas to match registered filings and production records to comply with regulations. Frequent data updates and diverse document types create multiple constraints. The update trigger for each batch or line change requires the knowledge base to support high-frequency incremental synchronization, to avoid retrieving expired data. Multiple document types include precise fields and semantic descriptions, requiring retrieval to support both exact matching and semantic recall. Core review links need to prioritize returning key information such as restricted substances and filed formulas. This requires the relevance ranking of recall results to be accurate, with no irrelevant documents interfering with the review process.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Cosmetic formulas and filing data require high matching accuracy to avoid mismatching raw materials or restricted substance information |
limit | Top 6–10 entries | The number of relevant documents for a single production batch is limited. Excessive results increase context processing burden |
searchMode | mixedRecall | Balances exact field matching for formula components and semantic retrieval needs for process descriptions |
usingReRank | Enabled | Re-ranks recall results by relevance to ensure core review content such as restricted substances and filing information is displayed first |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval scope to cover auxiliary information associated with the current batch in production records |
collectionIds | Collection IDs divided by business module | Manages formulas, production records, and packaging material data separately for precise retrieval of documents from corresponding modules |
Three Common Errors
- Symptom: A token count limit exceeded error is returned during retrieval, and no recall results can be generated. Cause: The token calculation rules for cosmetics production documents are not mastered, and chunking parameters are not adjusted based on average document length, causing the token count of a single post-chunk document to exceed the system preset upper limit.
- Symptom: No new index data is generated in the knowledge base after calling the training order creation interface. Cause: The functional difference between adding data to a collection and creating a training order is not clarified. Only text upload is completed, and the index construction process is not triggered.
- Symptom: Expired old packaging material documents are mixed in retrieval results. Cause: The incremental update trigger condition of the knowledge base is not configured according to the update frequency of each batch or line change, causing old data to not be overwritten in a timely manner.
How to Verify Configuration Correctness
- Upload a single typical cosmetic formula work order, run a retrieval test, and verify that the matching degree of returned results aligns with the preset
similaritythreshold. - Simulate a production scenario after a line change, trigger a knowledge base incremental update, and verify that the document update time of the specified collection is synchronized with the actual production batch.
- Call the retrieval interface to check whether the number of returned results matches the value set in the
limitconfiguration, to confirm that the parameter takes effect. - After enabling
usingReRank, verify that the sorting of returned results prioritizes documents related to core review fields such as restricted substances and filing information.
The values provided are common starting points and should be measured against the reader's own samples.