Data Characteristics for This Category
Tender listing pharmacovigilance data primarily originates from national and local centralized drug procurement platforms, public information from medical insurance bureaus, adverse reaction monitoring reports published by drug regulatory authorities, and procurement and usage records from various medical institutions. This data typically appears as structured or semi-structured documents, such as Excel spreadsheets, PDF announcements, XML files, or database export files. Data updates are frequent; some procurement information updates weekly or even daily, while adverse reaction reports are released monthly or quarterly. Documents include fields such as drug generic name, manufacturer, dosage form, specification, procurement price, winning bid provinces, adverse event type, occurrence time, and severity. Price units are typically in Chinese Yuan (RMB), and adverse reaction incidence rates may be expressed as percentages or per thousand persons.
Constraints Imposed by These Characteristics on "Deployment and Upgrade"
The high update frequency of tender listing data requires FastGPT deployments to have efficient data synchronization and incremental update capabilities to ensure knowledge base timeliness. Diverse document formats, especially PDFs and structured tables, challenge document parsing capabilities, necessitating the configuration of robust document parsers. Key fields within the data, such as drug generic name and manufacturer, require precise identification and extraction to support accurate pharmacovigilance queries and relational analysis. Specialized terminology and medical descriptions in adverse reaction reports demand high model comprehension and semantic matching accuracy. Additionally, numerical data like procurement prices and adverse reaction incidence rates must retain their numerical properties during knowledge base construction to enable range queries and statistical analysis.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
UPLOAD_FILE_MAX_SIZE | 500 MB | Tender announcements and adverse reaction reports may contain numerous attachments or images; this ensures large files upload successfully. |
maxContext | 3000 characters | Pharmacovigilance reports and tender specifications can be lengthy; this ensures sufficient context for semantic understanding. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing large PDFs or structured tables can be time-consuming; this prevents parsing timeouts. |
Chunk size | 800 characters | Balances semantic completeness and recall efficiency, avoiding excessive segmentation or overly long segments. |
Recall count | Top 10 entries | Ensures coverage of multiple relevant adverse reaction cases or tender details during pharmacovigilance queries. |
Similarity threshold | 0.75 | Improves recall accuracy, filtering out irrelevant query results and reducing noise. |
Common Pitfalls
- When creating a knowledge base, the general knowledge base fails to recognize image content because the deployment environment lacks support for multimodal models like
SenseVoiceSmall. - After deploying MCP Server locally, frequent service startup errors occur. This is typically due to
mcpserverproxyendpointbeing configured withlocalhost:Port, while the actual service listening address is not the loopback interface. - The drug price field is empty in query results. This usually means the document parser failed to correctly identify numerical price columns in PDF or Excel tables, leading to data extraction failure.
How to Verify Correct Configuration
- Upload a tender announcement PDF file containing images and tables. Check if the knowledge base correctly identifies and extracts image descriptions and table data.
- Attempt a query using a drug generic name. Verify that the recall results include relevant tender information and adverse reaction reports, and check if key fields (e.g., manufacturer, price) are complete.
- Perform a query involving a price range. Check if the system returns results matching the conditions based on numerical data, and verify that the numerical units are correct.
Note: The values provided are common starting points and should be measured against specific samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.