What Data for This Category Looks Like
Data related to medical device marketing primarily comes from internal enterprise marketing materials, regulatory public documents, and clinical research reports. Update frequency aligns with product iterations, regulatory policy adjustments, and marketing campaign updates. Most documents use mixed text and image layouts, including multi-level table of contents product manuals, structured parameter tables, clinical application descriptions, and promotional materials. Fields include registration certificate numbers, model specifications, applicable departments, clinical validation data, and units of measurement. Common units include millimeters, pascals, kilovolts, and similar units. Some documents contain embedded product diagrams and clinical case images.
Constraints Imposed on Document Parsing and Chunking
Multi-level table of contents document structures require the parsing module to accurately identify hierarchical relationships. Failure to do so will disrupt the logical connection between product parameters and descriptions. Documents with mixed text and images require retaining the binding relationship between parameter tables, product images, and their corresponding descriptions. Standard plain text parsing can easily lose this image-text association. Medical device documents contain a large number of structured parameter fields. Structured content must be extracted; plain text extraction alone cannot support precise retrieval. Additionally, document formats include Feishu collaborative documents, docx, PDF, and other types. Parsing compatibility across different formats requires configurations covering mainstream material formats, and OCR extraction for embedded images must be supported.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
PARSE_FILE_SUPPORT_FORMATS | ["docx", "pdf", "Feishu Docs", "md"] | Covers mainstream formats of medical device marketing documents, including commonly used Feishu collaborative documents and product manual formats |
maxChunkSize | 800–1200 characters | Balances contextual completeness for long clinical descriptions and structured parameters. Avoids broken parameter associations from overly small chunks, and reduced retrieval accuracy from overly large chunks |
PARSE_ENABLE_TABLE_EXTRACT | Enabled | Parameter tables in medical device documents contain core retrieval fields such as model and specification. Structured content extraction is required to support precise matching |
PARSE_IMAGE_OCR_ENABLE | Enabled | Medical device marketing documents often include product diagrams and clinical case images. OCR must be used to extract text within images to fully restore document content |
UPLOAD_FILE_MAX_SIZE | 500 MB | Matches the common file size of single medical device clinical research reports |
PARSE_TIMEOUT_SECONDS | 600 seconds | Prevents interruptions to large PDF or clinical report parsing due to excessive time consumption |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require individual analysis. It is recommended to test on your own samples before finalizing settings.
Three Common Misconfiguration Issues
- Issue: Only the first level of content is parsed from Feishu multi-level table of contents documents, with all other levels not retrieved. Cause: The full table of contents parsing switch for Feishu documents is not enabled. The default setting only parses the current page content.
- Issue: After calling the API to upload a doc format file, a successful response is returned but no parsing results are generated. Cause: The
docxformat is not configured inPARSE_FILE_SUPPORT_FORMATS, or the API request does not carry correct file parsing parameters. - Issue: The marker module reports split-related logs, and no parsing log output is seen for the docker-deployed v4.9.0 version. Cause: The parsing module's log directory is not mounted, or
PARSE_TIMEOUT_SECONDSis set too short, causing parsing interruptions that are not recorded.
How to Verify Correct Configuration
- Upload a medical device product manual with a multi-level table of contents, and check if all hierarchical table of contents content is included in the parsing results.
- Upload a doc file containing a parameter table and product images, and check if the parsing results extract the table content and text within the images.
- Call the API to upload a test file, and check if the returned results include parsed text fragments.
- View the parsing module's logs, confirm there are no split-related errors, and verify the parsing process completed normally.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.