What the data for this use case looks like
The core data for the campaign setup use case comes from advertising platform APIs, budget scheduling sheets, attribution tools, and targeting configuration documents. Daily processed documents include campaign plans, audience targeting settings, budget allocation rules, conversion tracking configurations, and more. Fields cover enumerated targeting tags, numeric budget amounts, temporal scheduling nodes, and other types, with units including yuan, impressions, days, and others. Data update frequencies are mostly daily or when a single campaign starts. A single document may contain multiple independent campaign unit configurations.
What constraints do these characteristics impose on document parsing and chunking
Campaign setup documents have a high degree of structuralization. Targeting, budget, and conversion information for a single campaign configuration are strongly correlated. Chunking must not split correlated content, otherwise subsequent Q&A will fail to fully match campaign units. A single document may contain multiple pages of budget schedules or multiple sets of targeting rules. When the file size is large, parsing time increases significantly, which easily triggers timeout errors. Document formats vary across different sources. Some documents exported from advertising platforms include fixed-format table fields. Parsing must retain field semantics to avoid loss of configuration information.
How to configure settings
This document is based on the FastGPT v4.17.0 configuration specification.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | The length of a single campaign configuration entry falls within this range, which preserves complete configuration units such as targeting and budget |
chunkSplitter | Semantic segmentation | The targeting, budget, and conversion configuration logic for campaigns is independent. Semantic segmentation avoids chunking across multiple configurations |
customPdfParse | Enabled | Adapts to standardized PDF document formats exported from advertising platforms, and supports structured field parsing |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Covers the parsing time required for most campaign setup documents with multiple scheduling pages, and avoids timeout errors for large files |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapts to batch upload requirements for campaign setup documents, and prevents files from being blocked due to size limits |
chunkSettingMode | Manual configuration | Allows adjustment of chunking parameters for different document formats, to adapt to structured content parsing |
Three common mistakes
- Phenomenon: Parsing tasks remain in a waiting state with no results after integrating
customPdfParse, and large file tasks occupy resources, causing subsequent tasks to block. Cause: No task queuing mechanism is configured, so large file parsing requests cannot be diverted. - Phenomenon: Calling
customPdfParseto parse a multi-hundred-page campaign setup PDF returns a timeout error. Cause: ThePARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted, and the default timeout duration is insufficient to cover the parsing time of large files. - Phenomenon: Parsed chunks contain a large amount of duplicate content, and the number of recalled campaign configurations cannot be controlled. Cause: Content deduplication and recall configuration are not enabled via
chunkSettingMode, resulting in redundant chunks and unexpected recall results.
How to verify that the configuration is correct
- Upload a campaign setup PDF containing multiple pages of budget schedules, and check whether the parsing task completes within the preset duration.
- Review the parsed chunk content, and confirm that a single chunk does not split complete information for a single campaign targeting or budget configuration.
- Upload a test file that exceeds the
UPLOAD_FILE_MAX_SIZEthreshold, and confirm that the system returns a file size limit exceeded prompt. - Adjust the chunking parameters, and verify that the number of parsed chunks and the number of recalled campaign configuration units meet expectations.
Note: The configuration values listed above are common starting points. They should be evaluated against the reader's own 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.