Knowledge Base Retrieval and Recall for Infectious Disease

Data Characteristics for This Category

Infectious disease registration and declaration data come from diverse sources. These include clinical trial reports, non-clinical study reports, epidemiological data, pharmacological and toxicological studies, manufacturing process documents, quality standards, risk management plans, and various guidelines and regulatory documents issued by regulatory agencies. Data updates occur frequently, especially for epidemiological data and clinical progress, with updates potentially happening quarterly or even monthly. Document structures vary, encompassing structured tabular data (e.g., clinical trial results), semi-structured text (e.g., research report abstracts), and unstructured long texts (e.g., expert consensus, literature reviews). Fields and units often involve microorganism names, infection sites, dosage units (e.g., mg/kg), treatment duration (days), efficacy indicators (e.g., cure rate percentage), and incidence of safety events. Standardization of microorganism names and infection sites is critical.

Constraints on "Knowledge Base Retrieval and Recall" from These Characteristics

Extensive data sources and high update frequency necessitate efficient data ingestion and update mechanisms for the knowledge base. This ensures the timeliness and accuracy of retrieval results. Diverse document structures require support for multi-type document parsing and effective extraction of key information from different structures. For example, the system must identify dosage and efficacy data from tables in clinical trial reports and extract key conclusions from lengthy reports. Standardization of specialized fields like microorganism names and infection sites demands higher requirements for query expansion and synonym matching to handle inconsistent terminology. Furthermore, frequent revisions of regulatory documents mean the knowledge base must distinguish between different versions during retrieval and prioritize recalling the latest and applicable regulatory provisions.

Configuration Guidelines

Configuration Item	Recommended Value	Rationale for Recommendation
`Chunk size`	`800–1200 characters`	Ensures individual segments contain sufficient context, preventing key information from being fragmented, while also balancing retrieval efficiency.
`Recall count`	`10–15 entries`	Covers a broader range of potentially relevant documents, especially when handling complex queries and multi-faceted information needs.
`Similarity threshold`	`0.75–0.85`	Balances recall rate and accuracy, reducing interference from irrelevant documents and improving the precision of retrieval results.
`Rerank result count`	`5 entries`	Focuses on the most relevant results, improving user reading and decision-making efficiency, and reducing information overload.
`UPLOAD_FILE_MAX_SIZE`	`500 MB`	Accommodates the upload requirements for large clinical trial reports and regulatory document packages.
`maxContext`	`8192 token`	Ensures the model can process longer queries and retrieved results for deeper understanding and synthesis.

Common Pitfalls

Retrieval results include a large number of outdated or currently inapplicable regulatory documents. This occurs because the knowledge base update strategy does not differentiate between versions and effective dates of regulatory documents.
Certain professional terms, such as specific microorganism names or infection sites, cannot be correctly matched and recalled. This is due to a lack of a specialized thesaurus and standardized terminology processing.
Complex queries submitted by users return too few results or results with low relevance. This happens because model context window limitations prevent a full understanding of the query intent, or the number of recalled items is set too low.

How to Verify Proper Configuration

Select queries of varying types and complexity. Verify that retrieval results include all expected key information.
Check the recall results for regulatory-related queries. Confirm that the latest and most relevant regulatory versions are prioritized.
Compare query results using standardized and non-standardized terminology. Evaluate the effectiveness of synonym expansion and term matching.
Simulate information needs in actual declaration scenarios. Evaluate the completeness and accuracy of recalled documents. Adjust Similarity threshold based on feedback.

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.