This page is written for FastGPT v4.17.0.
What Data for This Category Looks Like
Entities in this category’s data include slaughterhouses, cooked meat and frozen prepared food enterprises. Daily documents include quarantine certificates, slaughter and cutting records, ingredient marinating records, thermal processing records, cold chain temperature records, and microbial test reports. Data sources cover animal quarantine information, supplier batches, production traceability systems, temperature collection equipment, metal detection results, and laboratory test data. Updates run per production batch and shift. Cold chain temperatures are collected continuously. Document fields include batch number, quarantine number, temperature value, test indicators, qualification determination, and more. Units include batch numbers, degrees Celsius, test item names, and more.
How These Data Characteristics Impact Knowledge Base Retrieval and Recall
Frequent updates per production batch and shift, plus continuous cold chain temperature collection, require retrieval and recall to support precise filtering by batch number and time range. This prevents irrelevant data from other batches from being included. Documents contain highly compliance-focused fields such as quarantine and test information. Retrieval results must match required field completeness to meet compliance rules, and avoid missing key information. Field types vary widely across different documents: cold chain records use numeric values, while quarantine certificates use text. Retrieval and matching logic must adapt to multiple data types. Continuously collected temperature data has a large volume. It is necessary to control the scale of recalled results to avoid context overflow, which would reduce drafting efficiency for subsequent process documents and exception reports. Multiple document types associated with a single batch must be recalled in tandem to ensure consistent information across process documents and exception reports.
Configuration Settings
| Config Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets the precision requirement for matching highly compliance texts such as quarantine certificates and test reports, and avoids mixing low-relevance results |
limit | Top 8 entries | Adapts to the volume of documents associated with a single batch, controls context length, and avoids exceeding processing limits |
searchMode | embedding | Adapts to semantic matching for multiple field types, balancing numeric temperature records and text-based quarantine information |
usingReRank | Enabled | Optimizes result ranking in mixed multi-document scenarios, prioritizing recall of content directly related to processes and exceptions |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval scope based on batch numbers and quarantine numbers, associating multiple types of documents from the same batch |
collectionIds | Specified per business scenario | Only retrieves dedicated datasets containing quarantine and production records, avoiding interference from irrelevant data |
Three Common Misconfigurations
- The symptom is that uploaded quarantine certificates and microbial test reports display garbled text. The cause is failure to adapt to the encoding format of Chinese documents, or parsing failure due to encrypted fields in uploaded files.
- The symptom is an authentication failure prompt returned when calling the knowledge base API. The cause is using a universal KEY instead of a business-specific authentication credential, and failing to bind access permissions for the corresponding dataset.
- The symptom is that retrieved results include documents from multiple unrelated production batches. The cause is failure to configure batch number-based filtering rules, or failure to limit
collectionIdsto only retrieve the target dataset.
How to Verify Proper Configuration
- Upload a single quarantine certificate, check whether the retrieval result fully returns core fields such as quarantine number and qualification determination, with no garbled text.
- Enter a query containing a specific batch number, confirm that recalled results only include slaughter and cold chain records for the corresponding batch, with no cross-batch data.
- After enabling
usingReRank, compare result rankings before and after enabling, confirm that the ranking of highly relevant documents is optimized. - Check the dataset’s configuration logs, confirm that
datasetSearchUsingExtensionQueryhas triggered batch number-based extended retrieval.
The values provided here are common starting points and should be evaluated against local samples.