Data Characteristics
Telemedicine data sources are diverse. They primarily include regulations, rules, and guidelines from national and local health commissions. They also include internal telemedicine procedures, consultation standards, and medical record management details from healthcare institutions. Additionally, they cover medical insurance payment standards and data security/privacy protection agreements. Document update frequencies vary. National regulations might update annually or every few years. Internal SOPs might update quarterly due to technological advancements or service model adjustments. Document structures typically include chapters, sections, clauses, and appendices. Some documents embed charts and flowcharts. Common fields include "scope of application," "responsible party," "operation steps," "risk warning," and "data transmission standards." Units often include "hours," "working days," "MB/s," and "encryption algorithm name," representing technical or time measurements.
Constraints from Document Parsing and Chunking
Telemedicine policy documents have broad data sources and varied update cycles. This requires parsers to handle multiple file formats (PDF, DOCX, XLSX) and effectively identify version updates. The strict hierarchical structure of documents, such as "chapters" and "clauses," demands high logical integrity for chunking. This prevents splitting critical clauses. Fields like specific encryption algorithm names and transmission rates in "data transmission standards," and time or process nodes in "operation steps," require precise identification. They must remain within the same chunk to ensure RAG retrieval provides complete and accurate policy basis. Some documents may contain complex tables or flowcharts. Parsers must extract semantic information or convert non-text content into understandable text descriptions to avoid losing critical information.
Configuration Strategy
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
UPLOAD_FILE_MAX_SIZE | 100 MB | Telemedicine policy documents are often lengthy and contain multimedia content, requiring support for large file uploads. |
Chunk size (Chunk Length) | 800–1200 characters (characters) | Ensures each chunk contains sufficient context, covering complete clauses or operation steps, reducing semantic loss. |
Chunk Overlap Length (Chunk Overlap Length) | 100–200 characters (characters) | Ensures some contextual overlap between adjacent chunks, preventing critical information from being cut by chunk boundaries, improving retrieval coherence. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds (seconds) | Processing large PDF or complex DOCX files can be time-consuming, requiring ample time. |
File Collection Parsing Strategy | By Chapter/Title Structure | Telemedicine policy documents have clear hierarchies; parsing by structure preserves original logic, improving chunk quality. |
Enable Auxiliary Data | Yes | Utilizes document metadata (e.g., chapter titles, publication dates) as auxiliary information to enhance retrieval accuracy and relevance. |
Common Pitfalls
- RAG retrieval returns fewer Excel data entries than expected. This may occur because the parser did not correctly identify and convert some table structures or non-standard cell content into retrievable text chunks during Excel file parsing, preventing some data from entering the knowledge base.
- After uploading a PDF file, API requests do not return complete chapter titles. This typically happens due to poor PDF scan quality or missing text layers, causing Optical Character Recognition (OCR) to fail at accurately extracting title information.
- Auxiliary data in knowledge base chunks does not improve matching accuracy. This may be because auxiliary data (e.g., document type, source) was not correctly associated with the main content during indexing or was not considered in similarity calculations during retrieval.
Validation Steps
- Upload typical telemedicine policy documents (PDF, DOCX, XLSX). Check if the parsed chunk count and content meet expectations, especially verifying effective conversion of tables and flowcharts.
- Using the knowledge base retrieval interface, input specific clauses or keywords from policy documents. Check if the returned results include complete relevant chunks and verify their contextual coherence.
- For frequently updated documents, upload new versions. Check if the system correctly identifies and updates knowledge base content. Also, verify the retrieval of information from older versions.
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.