Data Characteristics
Infection control data originates from Hospital Information Systems (HIS), Laboratory Information Systems (LIS), Electronic Medical Records (EMR), adverse drug reaction monitoring systems, and various infection control reports. Data updates frequently. Some real-time monitoring data updates every minute. Weekly and monthly reports update periodically. Document types vary, including structured case reports, drug usage records, and microbial test results, as well as unstructured physician orders, nursing notes, and adverse event descriptions. Fields cover patient demographics, medication history, infection sites, pathogens, antimicrobial susceptibility, adverse reaction types and severity, and treatment measures. Units include dosage (mg, g), frequency (times/day), time (hours, days), and concentration (ug/mL).
Constraints on Knowledge Base Retrieval and Recall
High-frequency real-time data updates require the knowledge base to quickly synchronize and index data to ensure timely recall. Diverse and heterogeneous data structures necessitate detailed preprocessing and standardization during knowledge base construction. This is especially true for entity recognition and relationship extraction from unstructured text; otherwise, retrieval accuracy suffers. Colloquialisms mixed with medical terminology in adverse drug reaction descriptions demand advanced semantic understanding and fuzzy matching capabilities. Precise matching of critical fields like infection site, pathogen, and drug susceptibility is crucial for ensuring recalled content is highly relevant to actual infection control scenarios. Any deviation can lead to incorrect judgments or recommendations. Sensitive information in the data also imposes strict requirements on access control for retrieval results.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
Chunk size | 500–800 characters | Balances contextual completeness and retrieval efficiency; avoids noise from overly long paragraphs. |
Recall count | Top 5 entries | Balances relevance and processing load; ensures results focus on key information. |
Similarity threshold | 0.75–0.85 | Filters low-relevance results, addressing the precision requirements of medical terminology. |
Rerank result count | 3 entries | Further optimizes ranking to improve the accuracy of user-visible results. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Accommodates parsing time for large or complex structured reports. |
maxContext | 4096 tokens | Ensures large medical texts are fully understood and prevents information truncation. |
Common Misconfigurations
- Knowledge base retrieval returns no results, with the AI responding, "I cannot find relevant information." This occurs when the text segmentation strategy is too aggressive, splitting complete event descriptions and leading to incomplete information in individual segments that cannot be matched.
- Retrieved adverse reaction event information does not match the actual situation. This manifests as entity recognition errors or missing critical fields like time or dosage. This occurs when preprocessing rules for unstructured text do not adequately cover diverse medical expressions.
- Users cannot access or retrieve specific knowledge base content, resulting in a
403 Forbiddenerror. This occurs due to improper user authentication configuration on the knowledge base node, failing to correctly associate user roles with data access permissions.
Verification Steps
- Select multiple typical infection control cases with different pathogens, drugs, and adverse reaction types. Simulate user queries and check if the recalled results accurately include core entities and relevant event descriptions.
- Compare retrieval results with original documents. Verify that the recalled content is complete and untruncated, especially for numerical information like critical times and dosages.
- For documents containing sensitive information, test with user accounts having different permissions. Confirm that only authorized users can successfully retrieve and view the content.
- Monitor the knowledge base update frequency and compare it with the actual data source update frequency. Ensure the latest data is indexed and recalled promptly.
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.