Data Profile for This Category
The data for this category comes from pharmaceutical distribution enterprises’ pharmaceutical business license declaration materials, circulation traceability files, compliance self-inspection archives, and rectification documents from regulatory authorities. Data updates in real time during individual application processes, or archives and syncs once per quarter. Documents include structured form fields and unstructured attachments. Structured fields cover pharmaceutical generic names, business license numbers, and circulation batch numbers. Unstructured attachments are mostly scanned documents and signed, sealed files. Units used include batch, piece, hour, and other circulation-specific measurement terms.
Constraints Imposed on Knowledge Base Retrieval and Recall
The mixed structure of structured fields and unstructured attachments requires retrieval to support both precise keyword matching and semantic relevance recall. Precise identification fields such as circulation batch numbers and business license numbers need appropriate similarity thresholds to maintain recall accuracy. Real-time data updates tied to application processes require the retrieval data source to sync the latest archived files and declaration materials. This prevents expired or invalid content from being recalled. OCR recognition errors in unstructured scanned documents can break keyword matching. Additional secondary filtering logic is needed to address this.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Adapts to mixed matching needs for precise fields and semantic text in pharmaceutical distribution scenarios, avoids low-relevance results |
limit | Top 8–12 results | Covers the number of associated files and reference documents needed for a single declaration review, prevents context overload |
searchMode | hybrid | Combines keyword matching and semantic recall, supports mixed document types of structured fields and unstructured attachments |
usingReRank | Enabled, re-rank the top 5 recalled results | Performs secondary sorting on recalled results to filter low-quality matches from OCR recognition errors |
datasetSearchUsingExtensionQuery | Disabled | No extended query is required for declaration material and file review, avoids irrelevant general circulation industry documents |
collectionIds | Bind exclusive pharmaceutical distribution compliance dataset | Limits retrieval scope, excludes interference from other category documents, ensures targeted review results |
Three Common Misconfigurations
- Empty results during retrieval. Cause: The exclusive pharmaceutical distribution compliance dataset is not bound. The retrieval scope covers unrelated category documents, leading to no matching content.
- Garbled characters in the knowledge base for imported scanned attachments. Cause: The re-ranking function is not enabled. Error characters from OCR recognition are not filtered.
- Forced display of the knowledge base search result module in the answer interface. Cause: The default search result display configuration is not disabled for the review scenario. Unnecessary interface elements are added.
How to Verify Correct Configuration
- Access the dataset management page in the FastGPT v4.17.0 backend. Confirm that the exclusive pharmaceutical distribution compliance collectionIds are bound.
- Submit a test query with a pharmaceutical circulation batch number. Check that returned result similarity falls within the 0.72–0.85 range.
- Upload a scanned business license attachment. Verify no garbled characters appear in the knowledge base and that the re-ranking function works.
- Trigger a review query. Confirm the knowledge base search result module is not forcibly displayed in the interface.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.