What the data for this category looks like
Data for this category comes from advertising platform backends, customer qualification upload portals, contract management systems, and public regulatory rule repositories. Daily processed documents fall into four categories: advertising review forms, customer qualification attachments, delivery record ledgers, and rectification notice letters. Document structures vary significantly: advertising review forms contain structured table fields, qualification attachments are mostly stamped PDF files, delivery records are time-sorted text ledgers, and rectification notices are official letters with unique identification numbers. Data update frequency is adjusted based on individual advertising campaigns or calendar months. Core fields include delivery ID, qualification number, compliance review status, and others. This category focuses on retaining audit and delivery evidence, with a priority on evidence archiving and retrieval. Its focus differs from that of advertising material review document categories.
What constraints these characteristics impose on the "document parsing and chunking" link
Documents in this category originate from diverse sources and have large format differences. The parsing link must support structured tables, scanned PDFs, and plain text ledgers. Structured fields in advertising review forms must be fully preserved. Chunking must not split content that spans multiple fields, otherwise the connectivity of compliance evidence will be broken. Scanned qualification attachments may have text blocked by stamps. Parsing logic must be adapted to extract complete content. Data updates are tied to campaigns or calendar months. This requires support for fast parsing and chunking of incremental documents, to avoid reprocessing historical data. The time-series characteristics of delivery records require chunking to retain contextual connections. This ensures that complete delivery cycle evidence can be matched during retrieval.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Aligns with paragraph lengths of advertising review forms and single record lengths of delivery records, avoiding splitting critical compliance fields |
chunkSplitter | Split by paragraph + table | Compatible with structured tables in advertising review forms, retains field connections within tables, and adapts to paragraph logic for plain text delivery records |
customPdfParse | Enabled | Adapts to scanned PDFs of qualification attachments, extracts text blocked by stamps, and prevents parsing gaps. This configuration must be enabled in FastGPT v4.17.0 or later |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Adapts to the parsing duration of multi-page rectification notices, preventing timeout errors during batch uploads |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports uploading large-capacity customer qualification attachments, meeting the demand for batch qualification submissions for single campaigns |
Three common mistakes
- Partial text is missing when parsing PDF qualification attachments. Cause:
customPdfParseis not enabled. The default parsing logic cannot identify compliance text blocked by stamps. - Only single-column table data is extracted after importing structured tables from advertising review forms. Cause:
chunkSplitteris not configured to support multi-vector table splitting. The default chunking logic does not recognize table structures, leading to broken field connections. - Parsing timeout errors occur during batch uploads of rectification notices. Cause:
PARSE_FILE_TIMEOUT_SECONDSis not adjusted. The default timeout duration is insufficient for the parsing process of multi-page rectification notices.
How to confirm configurations are set correctly
- Upload a single PDF advertising review form with structured tables, and review the parsed chunked content. Confirm that all columns and rows of the table are fully extracted and correctly connected.
- Upload a scanned customer qualification attachment, and confirm that the parsed text includes all visible content in the attachment with no obvious gaps.
- Check the system parsing logs. Confirm that the parsing duration of a single rectification notice does not exceed 120 seconds, and there are no timeout error records.
- Upload a CSV-format delivery record exported from a spreadsheet, and confirm that the chunked content retains the connection between table column headers and corresponding data.
The values listed here are common starting points, and should be validated against samples appropriate to the target deployment context.
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.