Data Characteristics for This Category
High-value consumable pharmacovigilance data originates from healthcare institution reporting systems, manufacturer post-market surveillance reports, and regulatory public databases. Data update frequency is relatively low compared to pharmaceuticals, typically quarterly or annually. Updates occur ad-hoc for significant events like product recalls or instruction manual revisions. Document structures are often a mix of structured and semi-structured data, including fields such as product batch number, model, manufacturing date, expiration date, implantation or use date, patient basic information, adverse event description, and treatment measures. Units of measurement involve quantity (pieces, sets), dimensions (millimeters, centimeters), and weight (grams, milligrams). Dimension and model information are crucial for adverse event analysis.
Constraints Imposed by These Characteristics on "Deployment and Upgrade"
The low data update frequency for high-value consumables means incremental knowledge base updates do not need to be overly frequent. However, each update might involve a large volume of data, requiring high stability for data import and parsing. Semi-structured data, especially adverse event descriptions, demands robust text parsing capabilities to extract key information. The presence of specific fields like dimensions and models requires FastGPT to effectively differentiate between numerical and textual semantics during vectorization and retrieval, avoiding incorrect purely numerical matches. Furthermore, the timeliness of product batch and recall information necessitates advanced data version management and historical traceability features to ensure responses are based on the latest or specific historical data versions.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 2048 | Adverse event descriptions for high-value consumables are often lengthy, requiring a larger context window to capture complete information. |
Chunk size | 800–1200 characters | Balances the completeness of adverse event descriptions with vectorization efficiency, preventing excessive truncation of key information. |
Recall count | Top 8–12 entries | Ensures broader coverage of relevant adverse event reports and product documentation during retrieval, improving recall rate. |
Similarity threshold | 0.75–0.85 | Addresses the textual characteristics of high-value consumable adverse event descriptions, balancing retrieval precision and generalization. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Requires a longer parsing timeout when processing large PDFs or product manuals containing images. |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports uploading product manuals or batch reports containing numerous charts and detailed technical parameters. |
Three Common Mistakes
- After a knowledge base update, retrieval results do not include the latest recall information. This occurs because the data import script failed to correctly identify and extract batch recall date fields when processing semi-structured data.
- After upgrading the system to version
4.8.20, retrieval results for some high-value consumable models are empty. This might be due to changes in the vectorization model in the new version when processing specific product model field formats, requiring retraining or configuration adjustments. - When querying adverse events related to specific product dimensions, the system returns numerous irrelevant results. This happens because the text segmentation strategy does not differentiate between numbers and units, leading to dimension information being split or mismatched.
How to Verify Correct Configuration
- Upload the latest versions of product manuals and adverse event reports. Check if their content is fully parsed, especially key fields like model, batch, and dimensions.
- Perform targeted queries, such as "fracture events for XX model stent in
2023." Verify that the returned results cover all relevant adverse event reports and confirm the accuracy of the time range. - Simulate queries containing product dimensions (e.g., "catheter with a diameter of
3.0 mm"). Observe if the recall results precisely match documents containing that dimension description and exclude irrelevant numerical matches.
Note: The values provided are common starting points. Measure them against your 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 2026-09-21.