What the data for this category looks like
Coking coal data sources include coal mine production daily reports, port customs clearance documents, industry association supply and demand weekly reports, spot transaction ledgers, and professional research reports. Updates occur daily for spot price changes, weekly for port inventory data, and monthly for supply and demand balance sheets. Document formats include structured Excel transaction tables, PDF-format research reports and documents, and scanned paper delivery slips. Fields include professional indicators such as ash content, sulfur content, caking index, and colloidal layer thickness. Common units are percentage, g/g, millimeter, yuan/ton, and others. Some documents mix structured fields and unstructured explanatory text.
Constraints Imposed on Document Parsing and Chunking
Coking coal data has diverse sources, with documents mixing structured tables and unstructured text. Parsing workflows must support both formats. High-frequency spot and transaction data requires incremental parsing to reduce redundant processing. Professional indicator fields are closely linked. Chunking must retain cross-field context to avoid breaking indicator combinations. A significant share of documents are scanned slips. Text extraction requires OCR. Bulk transaction ledgers have large individual file sizes. Parsing must support resource configurations for large files.
How to Set Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
PARSE_OCR_ENABLE | Enabled | Coking coal documents include large numbers of scanned port documents and paper research reports. OCR is required to extract text content |
PARSE_TABLE_STRUCTURE | Retain original row and column structure | Coking coal transaction tables include professional fields such as ash content and caking index. Full structure retention is required to avoid loss of field associations |
CHUNK_SIZE | 800–1200 characters | Adapts to the information density per chunk of coking coal research reports and transaction data. Balances context completeness and retrieval accuracy |
CHUNK_OVERLAP | 100–150 characters | Retains cross-chunk associations of professional indicators. Prevents key indicators from being split into different chunks |
UPLOAD_FILE_MAX_SIZE | 2000 MB | Adapts to the maximum individual file size of coking coal bulk transaction ledgers. Prevents large file upload failures |
PARSE_TIMEOUT | 600 seconds | Covers parsing time for large scanned documents and bulk ledgers. Prevents mid-process timeout interruptions |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material format, data volume, and business rules. Specific issues require specific analysis. It is recommended to test on your own samples before finalizing settings.
Three Common Mistakes
- A scanned coking coal PDF is uploaded, and the parsing result has empty fields. The cause is that the
PARSE_OCR_ENABLEconfiguration is not enabled. Text content in scanned images cannot be recognized without this setting. - A coking coal transaction Excel file is uploaded, and the vectorization index result is abnormal. The cause is that the
PARSE_TABLE_STRUCTUREconfiguration is not enabled. Table fields are scattered into fragmented text, and field associations are lost. - A large coking coal transaction ledger file is uploaded via API, and a
413 Request Entity Too Largeerror is returned. The cause is that theUPLOAD_FILE_MAX_SIZEconfiguration is not adjusted to a value matching the file size.
How to Confirm Configurations Are Set Correctly
- A single scanned coking coal port document is uploaded. The parsing result is checked for extraction of professional indicators such as ash content and sulfur content. The OCR configuration is confirmed as active.
- A coking coal monthly transaction Excel ledger is uploaded. The parsed chunks are checked for retention of complete transaction fields and row and column structure. The structured parsing configuration is confirmed as enabled.
- Bulk coking coal supply and demand research report PDFs are uploaded. The parsed chunks are checked for splitting by logical units. The chunking parameter configuration is confirmed as matching the document information density.
- A large file is uploaded via the API. The returned status code is checked for
200 OKand absence of timeout prompts. The upload and parsing timeout configurations are confirmed as set correctly.
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.