What This Category's Data Looks Like
The primary data subjects for this category are slaughterhouses, cooked meat and frozen prepared food manufacturers. Daily documents include animal quarantine certificates, slaughter and cutting records, ingredient marinating records, thermal processing ledgers, cold chain temperature logs, microbial test reports, and more. Data sources cover animal quarantine information, supplier batch traceability, production process traceability data, real-time temperature collection, metal detection results, and laboratory test data.
Updates follow production batch and work shift schedules. Cold chain temperature data is updated continuously. Fields include batch number, quarantine number, temperature value, test indicator results, and more. Some documents require association with batch and quarantine information.
Constraints Imposed on Knowledge Base Retrieval and Recall
This category has strict data association requirements: batch and quarantine information must be one-to-one matched. Retrieval must accurately match the batch number field to avoid mixing cross-batch data.
Continuously collected cold chain temperature data has a high update frequency. Retrieval must support incremental recall and real-time index synchronization to prevent lagging data from interfering with results.
Document types include structured ledgers and unstructured test reports. Retrieval and recall must adapt to multi-format content.
Data anomalies can trigger compliance risks. Retrieval results must cover key compliance fields to ensure recalled content meets production traceability and compliance report requirements.
How to Configure
These configurations are based on the knowledge base retrieval feature of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | Threshold range 0.75–0.85 | Match precise fields such as quarantine number and batch number, to avoid low-confidence results mixing into compliance documents |
limit | Top 10 entries | Cover multiple ledger types for a single production batch, while controlling context length to adapt to report generation needs |
searchMode | embedding | Prioritize vector recall of core business fields for mixed structured and unstructured documents |
usingReRank | Enabled | Rerank key compliance data such as cold chain temperature and test results, to improve their recall priority |
datasetSearchUsingExtensionQuery | Enabled | Automatically expand synonymous retrieval for batch numbers and quarantine numbers, to cover different formats from handwritten and system-entered entries |
collectionIds | Configured per production batch group | Limit retrieval scope to the current work shift or batch's dataset, to avoid cross-batch data interfering with retrieval accuracy |
Three Common Configuration Mistakes
- Phenomenon: Retrieval result similarity values are abnormally high, exceeding the normal threshold range. Cause: No
similaritythreshold filter is configured, and original embedding output values are used directly. - Phenomenon: Some quarantine certificates and test reports cannot be retrieved after upload, with a prompt indicating no available embedding channel. Cause: No corresponding embedding channel is configured for the document dataset, or the uploaded file format exceeds the supported range.
- Phenomenon: A
Cannot convert undefined or null to objecterror is returned after the workflow calls knowledge base retrieval. Cause: The retrieval scopecollectionIdsis configured as empty, or no valid dataset ID is specified, resulting in empty results that trigger a type conversion error.
How to Confirm Configuration Is Successful
- Run a query with a known batch number, verify that returned results only include documents from the corresponding batch, to confirm the
collectionIdsconfiguration is effective. - Enter a query containing cold chain temperature keywords, check whether returned results include continuously collected temperature records, to verify that
usingReRanksorts results by business priority. - Upload a new quarantine certificate, wait for index completion, then retrieve the corresponding quarantine number, confirm that the document can be normally recalled, to verify the incremental index logic.
- Check the similarity values of retrieval results, confirm that results below the set threshold have been filtered, to verify that the
similarityconfiguration is effective.
The values given are common starting points and should be measured against the reader's own samples.