Reference Sourcing and Traceability for Structured Analysis

Data Characteristics

Infection control management data originates from hospital internal infection control reports, pathogen detection results, antibiotic usage records, infection control training materials, and relevant policies and regulations. This data updates frequently. For example, pathogen detection results may update daily, while policies and regulations release periodically based on national or local requirements. Document structures often include structured fields in reports, such as patient ID, infection site, pathogen name, and drug sensitivity results. However, they also contain extensive unstructured descriptions, like infection event narratives and intervention measures. Training materials and policies are primarily semi-structured text with chapter titles, lists, and charts. Fields and units involve microbiology nomenclature, drug dosage units (mg, g), time units (h, d), and percentage indicators like infection rates and incidence rates.

Constraints on Reference Sourcing and Traceability

High update frequency of infection control management data requires a knowledge base indexing mechanism that supports rapid incremental updates. This ensures the timeliness of reference sources. Mixed document structures (structured, semi-structured, unstructured) mean that text chunking must preserve the semantic integrity of different content types. For instance, a complete drug sensitivity result table cannot be split. Extensive specialized terminology and abbreviations (e.g., MRSA, VRE) demand higher accuracy from text embedding models. The model must correctly understand and recall relevant content. The reference traceability mechanism must precisely point to specific paragraphs or tables in original documents. This is critical for compliance and credibility, especially for policies, regulations, and medical guidelines. Additionally, documents containing sensitive patient information require de-identification during citation to ensure data security.

Configuration Guidelines

Configuration Item	Recommended Value	Rationale
`Chunk size` (Chunk Length)	`500–800 characters`	Balances short sentences in structured reports with the semantic integrity of unstructured descriptions. Prevents truncation of critical information.
`Recall count` (Recall Count)	`Top 8–12 entries`	Infection control management often requires multi-dimensional information. Increasing recall count improves relevance coverage.
`Similarity threshold` (Similarity Threshold)	`0.75`	The domain has many specialized terms. High similarity is required to avoid interference from irrelevant content.
`Rerank result count` (Reranked Return Count)	`Top 5 entries`	Reduces the amount of content presented to the AI, improves precision, and lowers model processing load.
`PARSE_FILE_TIMEOUT_SECONDS`	`600 seconds`	Allows sufficient parsing time for large PDF guidelines or reports with complex tables.
`maxContext`	`3000 tokens`	Ensures the AI has enough context to understand complex infection control event descriptions and multiple citation entries.

Common Pitfalls

Knowledge base retrieval results are empty or do not return references, providing only a generic response. This may be due to a Chunk size (chunk length) that is too small, leading to semantic fragmentation, or a Similarity threshold (similarity threshold) that is too high, filtering out relevant but not perfectly matching content.
The reference limit is set to 1500 characters, but actual reference content blocks exceed this limit. This usually occurs because the knowledge base chunking logic prioritizes semantic integrity over strict character count adherence, causing individual blocks to exceed the preset limit.
Reference data format does not meet expectations when knowledge base retrieval results are passed to subsequent HTTP requests or AI conversations in the workflow, causing processing failure. This happens when the quote field returned by the knowledge base is not correctly parsed or transformed in the intermediate steps.

Confirmation of Configuration

Test different queries for typical infection control issues. Check if returned references accurately point to key facts, data, or guideline sections in the original documents.
Examine parsing results for different document types in the knowledge base (e.g., infection reports, guidelines, policies). Ensure structured tables, lists, and other content are correctly identified and chunked.
Simulate frequently updated pathogen detection reports. Verify that the system can promptly update the index and accurately cite the latest data after new documents are added.
Review system logs to confirm that the PARSE_FILE_TIMEOUT_SECONDS parameter is sufficient to complete file parsing for large infection control guideline documents, without timeout errors.

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.