This document is written for FastGPT v4.17.0.
What the data for this category looks like
Common data sources for campaign performance optimization include advertising platform impression and click data, attribution system conversion data, CRM lead data, and landing page visit data. Commonly processed documents include daily campaign reports, weekly review reports, real-time bid adjustment records, and material A/B test data tables. Data update frequencies are mostly daily or real-time. Document structures are mostly a mix of structured tables and unstructured paragraphs. Fields include impressions, clicks, conversion cost, budget limit, and others. Units include yuan, count, percentage, and others.
What constraints do these characteristics impose on document parsing and chunking
Structured characteristics of advertising documents require that chunking does not break the row-column association of tables. Otherwise, subsequent Q&A will fail to match indicators and their corresponding values. Real-time or daily updated data requires fast parsing. Excessive parsing timeout will reduce campaign adjustment efficiency. Mixed documents from multiple sources (tables + paragraphs) need to adapt to different chunking rules. Generic chunking may split cross-type content. Daily reports and test tables uploaded in batches have large file sizes. Upload limits need to be relaxed while avoiding parsing errors triggered by oversized files.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the field length of advertising documents, avoids splitting associated data across table rows |
chunkSplitter | Mixed by table column + paragraph | Adapts to the mixed characteristics of structured tables and unstructured review content in daily documents |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Meets the volume requirements for batch uploading daily reports and material test tables |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the batch parsing time requirements of real-time updated data |
customPdfParse | Enabled | Adapts to structured table extraction for PDF format documents such as material test tables |
chunkSettingMode | Manual configuration | Allows adjustment of chunking rules for different document types, avoiding generic chunking from breaking data associations |
Three Common Mistakes
- Phenomenon: After uploading CSV-format bid adjustment records or test tables, only the first and second columns of data are displayed. Cause: Structured table parsing mode is not enabled. The default chunking logic only extracts plain text content and fails to recognize the association relationship of table column fields.
- Phenomenon: A 413 Request Entity Too Large error is triggered when uploading files after deployment in Docker mode. Cause: The
UPLOAD_FILE_MAX_SIZEparameter is not adjusted. The system default limit is smaller than the actual size of the uploaded file. - Phenomenon: The custom pdf-maker parser cannot properly extract table data from PDF documents. Cause: The parser key is not correctly set in the system configuration, or the
customPdfParseswitch is not enabled.
How to Confirm Configuration is Correct
- Upload a single PDF-format material test table with a volume not exceeding 1000 MB, and check whether the parsing result includes all column fields and their corresponding values.
- Submit batch structured table files, wait for parsing to complete, and confirm that there are no timeout error prompts.
- View the chunked text fragments, and confirm that no fragment splits associated data across table rows.
- Navigate to the system configuration page, and confirm that the
customPdfParseswitch is in the enabled state.
The values listed above are common starting points and should be evaluated against applicable 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.