What data for this category includes
Data sources include merged financial statements, standalone subsidiary audit reports, related party transaction filing documents, and equity penetration map documents. Updates are synchronized with quarterly financial reports, annual audits, and temporary related party matters. Most documents are Excel merged worksheets with multiple nested levels, or multi-page PDF files. Some documents are scanned PDFs. Fields include consolidated revenue, net asset value per individual entity, related party transaction amount, and similar items. Units are mostly ten thousand yuan or hundred million yuan. Some fields require cross-subsidiary summary calculations.
What constraints these characteristics impose on document parsing and chunking
Nested multi-sheet Excel documents require parsing tools to support traversal of all worksheets and data merging, to avoid missing subsidiary data. Scanned PDFs and long documents increase the complexity of OCR recognition and long text splitting. Cross-entity field associations require retaining document hierarchy and subject attribution information during chunking, to prevent data confusion during subsequent due diligence analysis. Frequently updated temporary announcement documents require parsing workflows to support incremental synchronization and format compatibility, to adapt to temporarily released non-fixed-format files.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
parse_multi_sheet | Enabled, traverse all data sheets and merge fields with identical names | Most due diligence documents for diversified holdings are multi-sheet consolidated reports, requiring coverage of structured data from all subsidiaries |
ocr_enable | Enabled, enable high-precision table OCR for scanned PDFs | Some due diligence reports are scanned documents, requiring full restoration of table and text content |
chunk_size | 800–1200 characters | Due diligence reports include nested data and long text. This range balances context integrity and chunk retrieval efficiency |
chunk_overlap | 150–200 characters | Cross-subsidiary associated fields require contextual continuity, preventing data association failures caused by chunk breaks |
parse_timeout | 300 seconds | Parsing long documents and multi-sheet data takes significant time, requiring sufficient processing time allocation |
enable_sheet_filter | Filter non-data sheets by document title | Due diligence reports often include non-core sheets such as annotations and tables of contents, reducing parsing and storage of invalid data |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by material format, data volume, and business rules. Specific issues require case-by-case analysis. It is recommended to test on your own samples before finalizing settings.
Three Common Mistakes
- Phenomenon: After uploading a multi-sheet Excel due diligence document, only data from the first sheet is indexed, and data from other sheets is missing. Cause: Multi-sheet parsing configuration is not enabled, or identical field merging rules are not configured.
- Phenomenon: After parsing a scanned PDF due diligence report, table content is garbled or key data is missing. Cause: High-precision OCR configuration is not enabled, or OCR parameters are not adapted to table layout logic.
- Phenomenon: After chunking a long document, cross-subsidiary related party transaction data is split into two independent chunks, making it impossible to obtain complete information via associated queries. Cause: Chunk overlap length is set too small, or document hierarchy and subject attribution information are not retained.
How to Verify Proper Configuration
- Upload an Excel due diligence template containing 2 or more data sheets, check if the parsing result includes structured data from all sheets.
- Upload a single-page scanned PDF due diligence report, view the parsed text preview to confirm table and text content matches the original document.
- View the chunk detail list, confirm that the long document is split evenly, and adjacent chunks contain the set overlapping content.
- Upload a temporary related party transaction announcement document, verify that the parsing workflow can correctly identify non-fixed-format temporary files.
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.