What the data for this category looks like
The core data for this category comes from advertising platforms, material libraries, user interaction data, and conversion data. The update frequency is typically daily or weekly. Commonly processed documents fall into four categories: creative scripts, material test sheets, delivery data reports, and iteration checklists. Creative scripts include fields for scene descriptions, delivery copy, and compliance prompts. Material test sheets include fields for test batches, material IDs, and click-through rate statistics. Delivery data reports include exposure volume, conversion volume, and cost data. Iteration checklists include creative adjustment records and compliance check items.
What constraints these characteristics impose on the "document parsing and chunking" link
The structural differences between the four types of documents require parsing adapted to different formats. The paragraph hierarchy of creative scripts, table associations in material test sheets, and long-text structure of delivery reports must all be fully preserved. The daily or weekly update frequency requires the parsing process to have sufficient stability to avoid interruptions during batch file processing. Fields related to advertisements and compliance prompt content must be prevented from being incorrectly split, as this could lead to missed compliance reviews or broken connections between creative-related information. Large-volume batch delivery reports and material collections require appropriate parsing thresholds to ensure processing efficiency.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the single-paragraph content length of creative scripts and delivery reports, avoiding splitting cross-field advertising creative copy and compliance prompts |
chunkSplitter | Split by paragraph + title | Matches the hierarchical structure of scenes, copy, and delivery instructions in creative scripts, preserving the association between table headers and data rows in material test sheets |
customPdfParse | Enabled | Adapts to the complex table structure of advertising material test sheets, preserving cell correspondence to avoid compliance review deviations caused by data misalignment |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Matches the parsing duration of daily updated delivery reports, avoiding interruptions during parsing of large-volume files |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Accommodates batches of creative scripts and material test sheet collections, meeting the upload requirements of weekly updated iteration checklists |
Three Common Errors
- After uploading a PDF, no parsing progress prompt appears, and search tests return empty results. The issue occurs because the
customPdfParseconfiguration is not enabled. Complex table-style material test sheets fail to parse correctly, leading to missing embedded content. - Parsing some PDF files results in empty content. This happens because the PDF contains encrypted layers or non-standard fonts, and the custom parsing mode is not triggered, causing text extraction to fail.
- Search results show mixed content across test batches. The issue occurs because the
chunkSizesetting is too small. It splits test data from the same batch into multiple chunks, resulting in lost contextual associations.
How to confirm the configuration is correct
- Upload a single creative script PDF, view the text preview after FastGPT v4.17.0 parses it, and confirm that the paragraph structure of scene descriptions and copy content is complete.
- Upload a material test sheet containing tables, and check whether the parsing result preserves the association between table headers and corresponding data rows.
- Upload a delivery data report of approximately 100 MB, wait for parsing to complete, and verify that no timeout errors occur.
- Trigger a search test, enter keywords related to a test batch, and confirm that the returned results include the accurate content of the corresponding document.
Additional Note
The configuration values listed above are common starting points, and should be evaluated against your own specific 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.