What data for this category looks like
Commercial property marketing content data sources include lease and sales plans for property owners, rental income calculation sheets, investment promotion activity plans, passenger flow statistics reports, and similar materials. Update frequency adjusts based on project milestones: high-frequency updates during investment preparation periods, and quarterly or semi-annual adjustments during operational phases.
Document formats include PDF, Word, and Excel, with significant structural differences. Some are complete project manuals combining text and images, while others are single-page activity notices. Excel spreadsheets often split multiple tables by building and business type.
Fields include shop number, gross floor area, internal area, rental unit price, business type, settled brands, and similar items. Common units are square meters, yuan/㎡/day, ten thousand yuan, and similar units.
Constraints imposed on document parsing and chunking
Mixed multi-format documents require the parsing process to support full format compatibility, to avoid missing parsing of non-PDF documents.
A high proportion of structured tables requires complete retention of row and column structures, to prevent conversion into chaotic text that affects subsequent analysis.
Documents combining text and images include materials such as floor plans and location maps. Text and original images must be distinguished to avoid OCR destroying original visual information.
Excel spreadsheets often split multiple sheets by building and business type. Support for extracting all sheet content is required, to avoid data omission caused by only extracting the first sheet.
Document length varies widely, from single-page short text to dozens of pages of long manuals. Adaptive adjustment of chunk granularity is needed to ensure contextual coherence.
How to set the configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
parse_mode | auto + retain_original_image | Commercial property documents often contain materials such as floor plans and location maps. Retaining original images avoids OCR destroying original visual information |
table_parse_strategy | full_structure | Documents include structured tables such as rental sheets and business type proportion sheets. Complete retention of row and column structures ensures subsequent retrieval accuracy |
chunk_size | 800–1200 characters | Covers chunking needs for short-text activity notices and long-text investment promotion plans, balancing contextual coherence and chunk granularity |
excel_parse_sheets | all | Commercial property Excel spreadsheets often split multiple sheets by building and business type. Extracting all sheets covers full business data |
parse_timeout | 300 seconds | Adapts to parsing needs for large investment promotion manuals with more than 50 pages, avoiding parsing interruptions caused by overly long documents |
ocr_strategy | only_text_blocks | Only triggers OCR for pure text blocks, retaining the original format of image materials, which meets business requirements for using original images |
The parameter values provided on this page are common recommended starting points for determining configurations. Actual values are affected by material form, data volume and business rules. Specific issues require specific analysis, and it is recommended to test on your own samples before finalizing settings.
Three common configuration errors
- Scenario: When importing marketing documents in Word/PPT format, after enabling
pdf_enhance, the parsing result does not retain the location map from the original document, and only displays OCR'd text descriptions. Cause: The PDF enhancement function only adapts to PDF format documents, and the original image retention logic is not triggered for non-PDF documents. - Scenario: When parsing an Excel business type distribution spreadsheet, the parsing result only includes content from the first sheet, and sheets corresponding to other buildings are not extracted. Cause:
excel_parse_sheetsis not set toall, and only the first sheet is parsed by default. - Scenario: When parsing an investment promotion manual containing a large number of structured tables, the table structure is lost in the output chunked content, and is converted into chaotic text paragraphs. Cause:
table_parse_strategyis not set tofull_structure, and tables are parsed as plain text by default.
How to verify correct configuration
- Upload a single PDF investment promotion manual containing floor plans, check whether the original embedded images are retained in the parsing result. Adjust the original image retention configuration based on business requirements for materials.
- Upload an Excel rental spreadsheet containing multiple sheets, check whether the parsing result includes content and corresponding sheet names from all sheets, confirming that the sheet extraction range meets business requirements.
- Upload a long-text Word marketing plan, check whether the chunked content is reasonably split within the configured
chunk_sizerange, with no excessive truncation or overly long chunks. - Upload a document containing structured tables, check whether the parsing result retains complete row and column structures, and is not converted into plain text paragraphs.
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.