What data for this category looks like
Data for this category primarily comes from the full production process of slaughterhouses, cooked meat and frozen prepared food enterprises. It includes documents such as animal quarantine certificates, slaughter and cutting records, ingredient marination records, thermal processing records, cold chain temperature records, and microbial test reports. Data is updated by production batch and shift, with real-time synchronization of continuous temperature control data. Document fields include batch number, quarantine number, temperature value, test results, and more. Some records must be bound to traceability information for corresponding raw materials and finished products. Units include degrees Celsius, batch numbers, test qualification identifiers, and more.
What constraints do these characteristics impose on the knowledge base retrieval and recall workflow
The batch binding feature of category data requires retrieval to accurately match batch numbers and quarantine numbers. This prevents cross-batch matching errors that could allow contaminated raw materials to enter production. The high-frequency updates of real-time temperature control data require the knowledge base to support incremental recall. This avoids retrieving expired temperature records that disrupt raw and cooked product zoning control. The strict control requirements for raw and cooked product zoning and sterilization temperature require retrieval results to simultaneously match process parameters and scenario tags. This ensures results align with current production links. The mixed storage of multi-source heterogeneous data (paper quarantine certificates, electronic temperature records) requires retrieval to support different document field formats. This improves recall matching accuracy.
How to set the configuration
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Matches strongly associated fields such as batch numbers and quarantine numbers, so the similarity threshold must cover the business need for precise matching |
limit | Top 8 entries | The number of documents related to a single batch of production is typically 5-7, retaining redundancy to avoid missing key records |
searchMode | fullTextRecall | Compatible with multi-source heterogeneous quarantine, temperature, and test documents, covering keyword matching across different fields |
usingReRank | Enabled | Reorders recalled results by production scenario priority, prioritizing core process parameters such as sterilization temperature and cold chain temperature control |
datasetSearchUsingExtensionQuery | Enabled | Automatically supplements extended queries for exclusive fields such as batch numbers and quarantine numbers, improving precise matching accuracy |
collectionIds | Bind corresponding knowledge base collections by production link | Differentiates documents from links such as slaughtering, marinating, and thermal processing, narrowing the retrieval scope to improve efficiency |
Three common configuration errors
- Symptom: Cannot retrieve batch numbers or quarantine numbers included in document titles, with critical information missing from recall results. Cause:
datasetSearchUsingExtensionQueryis not enabled, so the query logic for title fields is not automatically extended. - Symptom: Retrieval responses time out after deploying a high-parameter model. Cause: The
limitparameter is not set reasonably, with too many documents recalled leading to an overly long context that exceeds the model's processing limit. - Symptom: Irrelevant documents from raw and cooked product links are mixed in retrieval results, making it impossible to distinguish scenarios. Cause: Knowledge base collections corresponding to production links are not bound via
collectionIds, leading to an overly broad retrieval scope.
How to confirm the configuration is properly set
- Review the binding relationship of knowledge base collections, confirm that
collectionIdsis grouped by production links such as slaughtering, marinating, and thermal processing. - Initiate a test query that includes a batch number, verify that the
similarityscore of the recall results falls within the 0.72–0.80 range. - After enabling
usingReRank, verify that the recall results prioritize core process parameters such as sterilization temperature and cold chain temperature control. - Test retrieval of frequently updated temperature control documents, confirm that the recall results include the latest continuous temperature control records.
The values provided are common starting points and should be measured against the reader's own samples.