What this category of data looks like
Data for this category primarily comes from advertising delivery platforms, landing page links, product qualification documents, and advertising law compliance clauses. Commonly processed documents include ad creatives awaiting review, landing page screenshots, review opinion sheets, and modification records. Data updates occur daily or in real time. Most individual documents contain fields including creative identifiers, violation type tags, rectification requirements, and corresponding delivery accounts. No unified fixed units exist; fields and units are adjusted flexibly based on document type.
Constraints imposed on the document parsing and chunking link
Ad creative review data has scattered sources and frequent updates, so the parsing module must support synchronous access to multiple data source types. Landing page screenshots and ad creatives are mostly mixed-text and image unstructured documents, so the parsing and chunking logic must adapt to embedded image content. Review opinions and modification records are mostly short texts tied to specific creative identifiers. Context associations must be retained during chunking to avoid breaking the correspondence between creatives and rectification requirements. Real-time updated data requires the parsing module to have low-latency processing capabilities. Chunking rules for compliance clauses and business documents must also be differentiated to ensure the integrity of content related to violation determination.
Configuration settings
The currently compatible version is FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Ad creative copy and review opinions are mostly short paragraphs. This range preserves the complete semantics of a single violation determination or creative description |
chunkSplitter | Split by paragraph + title | Review documents are often structured by creative, violation item, and rectification opinion. Splitting by title preserves document structural relevance |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Parsing landing page screenshots and bulk multi-creative documents takes longer. This range covers the parsing duration of most conventional documents |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Bulk ad creative submission documents may contain multiple screenshots and text files. This upper limit meets conventional batch processing needs |
customPdfParse | Enabled | Advertising compliance clauses are mostly in PDF format. Enabling custom parsing preserves clause hierarchy and highlighted key content |
chunkSettingMode | Manual configuration | Chunking requirements vary significantly across different document types such as creative copy and compliance clauses. Manual configuration allows flexible parameter adjustments |
Three common mistakes
- Phenomenon: When parsing landing page screenshot documents, the returned results do not include content related to embedded images. Cause: Mixed image-text parsing configuration is not enabled, so image information is not included in the chunking process.
- Phenomenon: When entering a Yuque public share link as a data source, the system prompts that the URL cannot be parsed. Cause: Only public links in static web page format are supported. Yuque document share links must have no permission restrictions and match parsing rules.
- Phenomenon: When uploading multiple ad review documents in bulk, some parsed chunked content is missing. Cause: The
chunkSizeparameter is not adjusted to the range suitable for short-text review opinions, forcing long paragraphs to be split and breaking context associations.
How to verify correct configuration
- Upload an ad creative document that includes a landing page screenshot, and check if the parsed results include text descriptions or identifiers related to images.
- Enter a compliant public web page link, and confirm that the system can read it normally and generate chunked content.
- Upload a bulk packaged review document, and check if the parsing duration falls within the range of the preset timeout parameter.
- View the chunking configuration interface, and confirm that parameters such as
chunkSizeandchunkSplitterhave been manually adjusted according to business needs.
The values provided are common starting points and should be validated against sample data used in actual deployments.
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.