What the data for this category looks like
Data for this category primarily originates from advertising platforms, customer relationship management systems, attribution SDKs, and form collection systems. Daily produced documents include channel attribution rules, lead distribution tables, weekly channel reports, and exception reconciliation sheets. Data updates run on a daily or weekly schedule. Most document structures include fields such as channel identifier, lead source identifier, conversion time, and assigned sales ID. Some reports include numeric unit fields like advertising budget and conversion cost.
What constraints these characteristics impose on document parsing and chunking
Attribution rule documents are mostly structured configuration items. Field-to-field relationships must be preserved to support channel parameter verification. Lead distribution tables and reports are mostly multi-sheet Excel files. The business meaning of each sheet must be identified to complete duplicate lead deduplication and sales assignment confirmation. Exception reconciliation sheets are mostly semi-structured text. Normal data and exception markers must be distinguished to quickly locate budget mismatches. Since data sources are scattered, document formats include PDF, Word, Excel, and PPT. Parsing logic must be adapted for field extraction across different formats. Image-based channel identifiers must not be mistakenly processed via OCR. Chunk length must be controlled to retain context for cross-sheet lead distribution rules.
How to set configurations
The following configurations are based on the document parsing module of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the field association length of attribution rules and channel reports, avoids splitting cross-page channel parameter configurations |
chunkSplitter | Split by table and sheet | Adapts to the multi-sheet structure of lead distribution tables and segmented data descriptions in reports |
customPdfParse | Enable image retention mode | Prevents channel identifier images from being mistakenly processed via OCR, retains original images for subsequent channel verification |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the parsing duration of large weekly channel report files, avoids timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Accommodates large lead distribution Excel files with multiple sheets |
chunkSettingMode | Manual configuration | Adapts parsing rules for different document formats, avoids generic configurations damaging the field structure of attribution rules |
Three common configuration errors
- Enabling PDF enhancement results in images being processed via OCR, with no original images in the parsed output. This occurs because the image retention mode of
customPdfParseis not enabled, and full-document OCR is enabled by default. - Unable to read sheet names in Excel files. This occurs because
chunkSplitteris not configured to split by sheet mode, and only the first sheet is parsed by default. - No parsing results when importing PPT or Word documents. This occurs because the document format is not in the supported list, or the corresponding format parsing switch is not enabled.
How to confirm configurations are correctly applied
- A PDF document with a channel identifier image may be uploaded. The parsed output is checked to confirm if the original image is retained, verifying the
customPdfParseconfiguration is active. - An Excel lead distribution table with multiple sheets may be uploaded. The parsed output is checked to confirm if content and names of all sheets are included, verifying the
chunkSplitterconfiguration is correct. - A weekly channel report under 1000 MB may be uploaded. Parsing is allowed to complete, and confirmation is made that no timeout errors are triggered.
- Attribution rule documents in PPT or Word format may be uploaded. The parsed output is checked to confirm if structured configuration items are extracted from the document, verifying format support is active.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.