Data Profile for This Category
The primary entities covered by this category are liquid milk, yogurt, milk powder, and dairy ingredient producers. Daily documents include raw milk acceptance records, standardized batching sheets, pasteurization or UHT processing records, CIP cleaning records, filling records, sample retention reports, and microbial inspection reports. Data sources cover raw milk testing systems, MES manufacturing execution systems, homogenization and heat treatment equipment, online temperature and flow collection systems, CIP cleaning systems, and LIMS laboratory management systems. Updates are triggered per milk tank, per production batch, or in real time. Document fields include testing indicators, process parameters, and equipment operation data, with clear metrological requirements for corresponding testing items and parameters.
Constraints on Knowledge Base Retrieval and Recall
The high-frequency real-time update feature of dairy processing requires knowledge base retrieval to quickly respond to queries linked to a single production batch, to prevent delays that disrupt process verification. Multi-source heterogeneous data types require unified field mapping rules to avoid matching deviations across documents from different devices and systems. Process compliance validation for perishable raw materials requires retrieval results to prioritize the latest batch data, to ensure parameter timeliness. Differences in fields across documents—such as antibiotic indicators in raw milk acceptance records and aseptic parameters in filling records—require targeted recall matching rules to prevent cross-category field confusion. Process errors can lead to serious consequences, so retrieval result accuracy must be maintained to reduce false recall risks.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Basis for This Setting |
|---|---|---|
similarity | 0.75–0.85 | Dairy processing process parameters require high matching accuracy to avoid recalling irrelevant cross-batch data. This range balances recall rate and accuracy |
limit | Top 8 entries | The number of documents associated with a single production batch is moderate. Recalling 8 entries covers core scenarios such as process verification and cleaning records, and avoids context overload |
searchMode | mixedRecall | Combines keyword and semantic retrieval, adapting to document types that have both fixed parameters (such as pasteurization temperature) and unstructured descriptions (such as microbial test report abnormalities) |
usingReRank | Enabled | For multi-source heterogeneous data, re-ranking filters low-correlation results and ensures the recall priority of key parameters such as heat treatment temperature and time and CIP verification |
datasetSearchUsingExtensionQuery | Enabled | Expands the original query into process-related associated queries, for example, expanding "raw milk acceptance standards" to "raw milk antibiotic detection thresholds, physical and chemical indicator requirements", adapting to multi-dimensional process queries in dairy processing |
collectionIds | Configured grouped by document type | Raw milk acceptance records, CIP cleaning records, and filling records correspond to independent datasets, which can accurately limit the retrieval scope and avoid mismatches across document types |
Three Common Configuration Errors
- After calling the API to upload PDF-format process records, the knowledge base cannot parse structured parameters within tables. Cause: PDF enhanced parsing configuration is not enabled, and only plain text is extracted, resulting in lost table fields.
- After importing Excel-format raw milk acceptance records, some testing indicator fields appear empty. Cause: The Excel file contains merged cells or non-standard column names, and structured preprocessing has not been completed.
- Knowledge base retrieval fails to retrieve storage path information for embedded images. Cause: Image index extraction rules are not configured, only document text content is extracted, and image storage paths are not associated.
How to Verify Proper Configuration
- Upload a standard raw milk acceptance record PDF, call the retrieval interface, and confirm that the antibiotic detection threshold data in the document is recalled.
- Import an Excel batching sheet with merged cells, and check that all fields in the knowledge base are complete with no empty values.
- Submit a query for "pasteurization temperature parameters", and confirm that the re-ranking function prioritizes the latest batch of heat treatment records.
- Call the file deletion interface, and confirm that uploaded test documents can be removed normally.
The values provided here are common starting points and should be evaluated against samples from the actual use case.