What this category of data looks like
This category of data originates from advertising platforms, landing pages, product qualifications, and advertising law-related clauses. Commonly processed documents include ad creative copy, landing page screenshots, review opinion sheets, and creative modification records. Data updates occur daily or in real time. Document structures include standardized fields such as creative copy content, violation type identifiers, and review timestamps. Some documents like landing page screenshots use unstructured image formats. Most review opinion sheets are text formats with annotations.
Constraints imposed by these characteristics on document parsing and chunking
Ad creative copy is often long text that includes brand language and compliance reminders. Avoid splitting continuous compliance-related content during chunking. Landing page screenshots are unstructured images. Perform OCR parsing first before chunking, and retain the association between images and their corresponding copy. Review opinion sheets include annotations and numbers. Retain contextual connections during chunking to ensure accurate violation point localization. Real-time updated data requires controlled chunk processing duration. Prevent delays that disrupt review workflows. Data from multiple sources has significant format differences. Adapt parsing rules for different document types to ensure complete field extraction.
Configuration Settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Ad creatives and review opinions are mostly medium-length text. Avoid splitting continuous compliance-related content |
chunkSplitter | Split by semantic paragraphs | Ad creatives include structured content such as brand descriptions and compliance clauses. Splitting by paragraphs retains contextual connections |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | OCR parsing for landing page screenshots and batch processing of multiple documents require sufficient duration to prevent timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Single ad creatives may include multiple high-definition landing page screenshots. Support large file uploads |
customPdfParse | Enabled | Most review opinion sheets are PDF documents with annotations. Custom parsing retains annotations and original formatting |
chunkSettingMode | Manual configuration | Document formats vary significantly across sources. Manual configuration adapts to different chunking requirements after processing ad creatives and screenshots |
Three Common Mistakes
- Phenomenon: Fields in multi-column ad creative review tables are misaligned and content is chaotic after parsing. Cause: Default chunking rules do not adapt to multi-column text structures, and corresponding parsing configurations are not enabled.
- Phenomenon: After uploading an ad creative PDF with embedded images, the parsed result only contains text with no associated image information. Cause: The
customPdfParseconfiguration is not enabled. The default parsing does not extract embedded image content. - Phenomenon: When parsing interface log documents exported from advertising platforms, Set-Cookie fields are truncated or missing. Cause: Default chunking rules split continuous HTTP response header content, and complete fields are not retained.
How to Verify Proper Configuration
- Upload a single review opinion document with multi-column tables. Check if parsed chunked content retains column associations. Adjust chunking rules until the expected result is achieved.
- Upload an ad creative PDF document with embedded images. Check if the parsed result includes association information between images and their corresponding text. Confirm that the
customPdfParseconfiguration is active. - Batch upload 10 landing page screenshot documents. Monitor processing duration. Adjust the
PARSE_FILE_TIMEOUT_SECONDSvalue until no timeout errors occur. - Upload an interface log document exported from an advertising platform. Check if Set-Cookie fields are fully retained in chunked content. Confirm that the chunking rule adapts to HTTP response header formats.
The values listed here are common starting points. They should be evaluated against relevant 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.