What this category's data looks like
Data for this category comes from multiple business documents of skincare, haircare, makeup, and special cosmetics manufacturers. These documents include formula process sheets, weighing and feeding records, emulsification and filling records, cleaning and sanitizing records, semi-finished and finished product inspection reports, and label review documents. Data sources cover registered and filed formulas, raw material safety materials, production equipment parameters, ERP/MES systems, laboratory test results, and packaging material version information. Updates are triggered per production batch or each line change. Documents include fields such as raw material components, process parameters, inspection indicators, and packaging material claims. Some documents require matching unique identifiers like filing numbers and production batch numbers.
What constraints this category's characteristics impose on knowledge base retrieval and recall
The multi-source heterogeneous nature of this category's data requires retrieval systems to support cross-document field mapping. It must align unified fields across registered formulas, production records, and inspection reports to avoid retrieval ambiguity. The high-frequency update rhythm per batch or line change requires the knowledge base incremental synchronization mechanism to adapt to production cycles. This prevents data lag from impacting real-time Q&A. Business scenarios that require matching unique identifiers like filing numbers and production batch numbers require retrieval logic to link primary key fields. This avoids cross-batch confusion. Precision requirements for process parameters and inspection indicators require recall results to strictly align with original document fields. This prevents parameter deviations caused by fuzzy matching.
How to set configurations
These configurations are adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Cosmetics manufacturing documents require precise matching of process parameters and filing information. This range filters low-relevance results and avoids recalling unrelated formulas |
limit | Top 8 entries | The number of documents associated with a single production batch is limited. Recalling too many entries exceeds the context window. Recalling too few fails to cover full-process information |
searchMode | mixedRecall | Combines keyword and semantic retrieval. This adapts to content with both clear keywords and semantic understanding needs, such as formula components and process steps |
usingReRank | Enabled | Recalled results must be reordered by relevance. This ensures core process parameters and filing information are returned first |
datasetSearchUsingExtensionQuery | Enabled | Retrieval terms can be extended based on user questions. This adapts to parameter queries with vague descriptions in process procedures. For example, "emulsification temperature control range" is extended to "emulsification process temperature parameters" |
collectionIds | Bind corresponding knowledge base collections by business line | Distinguishes document collections for skincare, makeup, and other categories. This avoids cross-category retrieval confusion and matches the multi-category characteristics of cosmetics manufacturing entities |
Three common mistakes
- A "no matching content found" error is returned during retrieval. This occurs when the corresponding cosmetics manufacturing knowledge base collection
collectionIdsis not correctly bound, or when documents such as registered formulas and production records are not imported into the specified collection. - Correct knowledge base content does not appear in the response body. This occurs when
usingReRankis not enabled, or when thesimilaritythreshold is set too high. This filters relevant results before they can be included in the context window. - Knowledge base reference content is not returned during API calls. This occurs when the associated knowledge base collection ID is not specified in the request parameters, or when
datasetSearchUsingExtensionQueryis not enabled. This prevents retrieval terms from covering business scenarios.
How to confirm the configuration is properly set
- Import a single cosmetics formula process sheet, perform a targeted retrieval, and check whether the recall results include the core fields of the corresponding document.
- Adjust the
similarityparameter range, and verify that the effect of filtering low-relevance results meets business expectations. - Initiate a knowledge base Q&A request via the API, and confirm that the returned results include references to associated knowledge base documents.
- Simulate a production query after a line change, and verify that the knowledge base can recall cleaning and feeding records for the corresponding batch.
The values provided are common starting points and should be evaluated against the reader's own samples.