Document Parsing and Chunking for Coke Marketing Content

Marketing and industry data related to coke primarily comes from supply and demand documents released by industry associations, quote files from spot

What Data for This Category Looks Like

Marketing and industry data related to coke primarily comes from supply and demand documents released by industry associations, quote files from spot trading platforms, procurement demand letters from steel mills, company-made marketing manuals, and futures delivery standard documents. Update rhythms vary significantly: spot documents are updated daily, industry research reports are released weekly or monthly, and marketing materials are updated irregularly as needed. Document formats include structured numerical tables and text analysis paragraphs. Some are converted scanned paper reports, and some documents contain nested formulas and charts. Fields covered include origin, quote, particle size range, strength indicators, and supply and demand data, which correspond to name, currency unit, length unit, standard test score, and weight unit respectively.

Constraints on Document Parsing and Chunking

The data sources for the coke category cover multiple formats and have distinct update rhythms, which create multiple constraints for document parsing and chunking. Structured indicator tables contain professional fields such as particle size and strength. When chunking, the correlation between fields must be retained, and content spanning multiple related fields must not be split. Scanned paper research reports and marketing manuals require OCR preprocessing first; otherwise, valid text content cannot be extracted. Daily updated spot quote documents have compact content, so chunk length must be adapted to short text scenarios. Long-cycle industry research reports must be split by chapter to retain complete analysis logic. Some documents contain nested formulas and charts. The parsing link must retain the structural integrity of the original data to prevent chunking from destroying the readability of professional content.

Configuration Settings

Configuration ItemRecommended ValueRationale
ocr_enabledEnable for scanned documents, disable for non-scanned documentsScanned coke documents cannot extract text directly. OCR must be used to convert them into editable content before processing
max_chunk_size800–1200 charactersCoke marketing content includes both short-text quotes and long-text industry analysis. This range balances the integrity of both types of content and avoids splitting professional indicator combinations
chunk_overlap100 charactersCoke’s professional indicators and analysis content often connect across paragraphs. Overlapping chunks prevent key information from being split between two chunks
keep_table_structureEnableCoke’s indicator tables contain associated fields. Retaining table structure ensures field relevance during retrieval and improves retrieval accuracy
parse_timeout300 secondsOCR and parsing for large scanned research reports and marketing materials require a long duration. This timeframe covers processing for most files
max_file_size500 MBCovers the single-file size threshold for most coke industry documents and marketing materials

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 individual analysis. It is recommended to test on your own samples before finalizing settings.

Three Common Mistakes

  • Symptom: After uploading a scanned coke document, the knowledge base search returns no results. Cause: The ocr_enabled configuration item is not enabled, so valid text content in the scanned document cannot be extracted.
  • Symptom: After configuring chunking parameters, knowledge base search fails to recall target coke quote content. Cause: The chunk length is set outside a reasonable range, or the chunk overlap length is insufficient, causing key indicators and associated content to be split between different chunks.
  • Symptom: Parsed chunks show breaks in professional fields. For example, coke’s strength indicators and corresponding origin information are split into two chunks. Cause: The keep_table_structure configuration is not enabled, or the chunk length is set too small, destroying the field relevance within the table.

How to Verify Proper Configuration

  • Upload a test scanned coke document. Verify that the parsed text content includes the indicators and text from the original document to confirm whether the OCR configuration is effective.
  • Upload a coke marketing manual containing structured tables. Verify that the parsed chunks retain the original table structure to confirm whether the keep_table_structure configuration is enabled.
  • Adjust the max_chunk_size and chunk_overlap parameters. Manually check the chunk preview interface to confirm that professional indicators and associated content are not split into different chunks.
  • Submit a knowledge base search test. Enter professional terms or specific indicators from the document, confirm that the corresponding chunk content can be recalled to verify the rationality of the chunking configuration.

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.