What the data for this category looks like
Typical entities in this category are bread, pastry, biscuit, instant noodle, and frozen pastry factories. Daily documents include formulas 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, ovens or frying lines, laboratories, and food trace ledgers. Update frequencies follow per batch, per shift, or daily cycles. Document fields include allergen labels, baking/steaming temperature and duration, feeding weight, shelf life, and some documents have section titles and internal links.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
The high-frequency update nature of this category requires the retrieval pipeline to support real-time or near-real-time data synchronization. This prevents returning expired formulas or cleaning records. The multi-source document structure requires retrieval to be limited to the exclusive set of processing procedures and technical standards. This avoids mixing in non-target documents such as factory inspection reports. Frequent product line change requirements mean retrieval must accurately match the process parameters of the corresponding product. For example, proofing time for whole wheat bread and baking window for biscuits must not be confused. High attention to allergen-related fields requires retrieval results to prioritize content with allergen labels. This prevents mismatches that could cause risks.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Filter low-relevance non-process documents, ensure returned content highly matches user questions about process parameters |
limit | Top 5 entries | The number of process documents for a single production batch is limited. Limiting the number of results avoids redundant information interfering with production decisions |
searchMode | mixedRecall | The query examples combine domain terms with natural-language context; compare with representative data, not as a product default. |
usingReRank | Enabled | Perform secondary ranking on initial retrieval results, optimize the display order of core fields such as allergens and baking parameters |
datasetSearchUsingExtensionQuery | Enabled | Support converting colloquial user questions into standard process terminology. For example, "how long to bake" matches the duration parameter in baking curves |
collectionIds | Specify exclusive process procedure collection ID | Limit retrieval scope, exclude non-target documents such as cleaning and disinfection records and factory inspection reports |
Three common mistakes
- Phenomenon: Retrieval results are split into multiple independent document fragments, and a complete set of associated process information cannot be returned. Cause: The
collectionIdsparameter is not limited to the exclusive process collection, or a reasonablelimitparameter is not set, leading to redundant results interfering with production decisions. - Phenomenon: No document citation identifier is displayed at the end of the answer paragraph. Cause: The knowledge base's citation display function is not enabled, or the FastGPT version is lower than v4.9.7.
- Phenomenon: Retrieval timeout or empty results occur in high-concurrency scenarios. Cause: The caching policy is not adjusted based on the daily or per-batch data update frequency, or the number of single retrieval results is not limited, leading to excessive resource occupation.
How to confirm the configuration is correctly set
- Upload a test frozen pastry production document, retrieve "steaming parameters for frozen dumplings", check if the returned results include the corresponding steaming curve and feeding list content.
- Check if a corresponding document section citation identifier is displayed at the end of the answer, confirm that the citation function is operating properly.
- Initiate multiple concurrent retrieval requests, check for abnormal errors or missing results, verify concurrency adaptability.
- Use colloquial questions for testing, such as "allergen alerts for whole wheat bread", check if the corresponding formula document is accurately matched.
- Use FastGPT v4.17.0 or a later version to ensure all configuration functions work properly.
The configuration values listed above are common starting points. They should be evaluated against the specific samples and use cases of the target deployment.