Data Characteristics
Infectious disease product data comes from various sources, including clinical trial reports, drug inserts, CDC epidemiological data, academic papers, and adverse drug reaction reports. This data updates frequently, especially epidemiological data and new pathogen research, with updates potentially occurring weekly or even daily. Document structures typically include precise medical terminology and data, such as disease definitions, pathogen information, transmission routes, diagnostic criteria, treatment plans, drug dosages and side effects, and vaccination guidelines. Common fields include ICD-10 codes, CAS numbers, ATC classification codes, R0 values, and MIC values. Units strictly follow international standards, such as mg/kg, IU/mL, ℃, and ng/dL.
Constraints on Knowledge Base Retrieval and Recall
The high update frequency of infectious disease product data requires the knowledge base to support rapid synchronization and incremental updates. This prevents the recall of outdated or inaccurate information. The precise medical terminology and specialized coding systems demand higher accuracy from text segmentation and vectorization models. Models must effectively identify and differentiate synonyms, near-synonyms, and polysemous words within medical contexts. The coexistence of structured and unstructured data means that a single text retrieval strategy is insufficient. Entity recognition and relationship extraction techniques are necessary to precisely locate key information from complex clinical reports. Furthermore, queries for numerical data like drug dosages and pathogen toxicity require the knowledge base to support both text retrieval and numerical range queries with unit conversion.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
Chunk size (Segment Length) | 500–800 characters | Balances contextual completeness with retrieval granularity. This avoids diluting key information in long paragraphs while maintaining the coherence of medical concepts. |
Recall count (Number of Retrieved Items) | Top 8–12 items | Addresses complex queries that may involve multiple related knowledge points. This increases recall diversity and improves coverage. |
Similarity threshold (Similarity Threshold) | 0.75–0.85 | The field of infectious diseases demands high precision in terminology. A higher similarity ensures retrieval accuracy and reduces false positives. |
Rerank result count (Number of Reranked Items) | Top 5 items | While ensuring high recall, reranking focuses on the most relevant top items, improving the precision of the final presentation. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Accommodates the longer parsing times for large clinical trial reports or guideline documents. This prevents parsing failures due to timeouts. |
maxContext | 32000 tokens | Ensures the large language model can process longer medical contexts. This prevents truncation from omitting critical diagnostic or treatment information. |
Common Pitfalls
- Garbled characters after uploading files to the knowledge base: The file encoding format does not match the system's default encoding, or the file itself is corrupted.
- Irrelevant historical data in recall results: The knowledge base has not been incrementally updated or old data has not been cleaned. This leads to the recall of obsolete diagnostic and treatment plans.
- Large language model fails to cite original database excerpts in answers:
Function CALLis not correctly configured withtool_code, or the returnedtool_outputlacks identifiers for the original data source.
Verification Steps
- Select complex queries for typical infectious diseases (e.g., influenza, pneumonia) such as diagnostic criteria and treatment plans. Verify that the recall results include all key information and assess its timeliness.
- Upload documents containing special medical symbols or complex tables. Check that the knowledge base parses the content completely and without garbled characters, and that field extraction is accurate.
- For numerical queries, such as drug dosages or R0 values, verify that the recall results correctly identify numerical ranges and provide contextual information with corresponding units.
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 2026-09-21.