What this category of data looks like
This data mainly originates from the national petition information system, local petition systems, the 12345 citizen hotline platform, grassroots governance platforms, and historical petition archives, with daily updates. Commonly processed documents include petition registration forms, acceptance notices, transfer and assignment letters, handling opinions, review and re-review applications, resolution plans, and more. Most documents contain fields such as requesting party, matter details, handling process nodes, and policy basis, with no unified fixed format. Some paper scanned documents require OCR for text recognition.
Constraints on Document Parsing and Chunking
Diverse data sources create inconsistent document formats, with a mix of scanned and electronic documents. This requires adaptive multi-source document parsing capabilities. Daily updated batch documents demand stable parsing efficiency to avoid timeouts and interruptions. Significant field differences across documents require retaining key information such as handling processes and policy basis during chunking, to avoid splitting that breaks business logic integrity. Historical archives may use outdated formats, requiring compatibility with low-version document formats. It is also necessary to accurately identify associated content between requests and handling links, to provide accurately segmented business data for subsequent review links.
Configuration Settings
These configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Petition-related documents contain long paragraphs of policy basis and matter details. This range preserves business logic integrity and avoids splitting critical information |
chunkSplitter | Split by paragraph + title | Petition documents often have clear handling process titles. Splitting by paragraph + title preserves node association, which helps subsequent review quickly locate content |
customPdfParse | Enable pdf marker high-precision mode | Some petition documents are paper-scanned archived items. High-precision mode reduces OCR error rate and improves text recognition accuracy for scanned documents |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Reserve sufficient parsing time when processing large case files in batches, to avoid interrupting the entire review process due to timeout |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Compatibility with single large case file archives, meeting the upload and parsing needs of batch archived documents |
chunkSettingMode | Manual configuration | Flexibly adjust chunking rules for different types of petition documents to adapt to diverse business scenarios |
Common Mistakes
- OCR Error is displayed on the interface, and parsed text lacks scanned content. Cause: Failed to enable
customPdfParsehigh-precision mode, or OCR process is interrupted due to insufficient parsing service resources. - Upload a large case file and receive a parsing timeout prompt, but the FastGPT backend shows the parsing service completed successfully. Cause: The
PARSE_FILE_TIMEOUT_SECONDSconfiguration value is too low, not matching the actual parsing time of large documents. - Upload an Excel-format petition statistical file in a workflow, and cannot read the content. Cause: Corresponding file parsing rules are not configured, or file reading permission of the workflow node is not enabled.
- Some models do not trigger file parsing actions. Cause: Preconditions for triggering parsing are not set, or model context length limits prevent recognition of uploaded file content.
How to Verify Proper Configuration
- Upload a petition scanned document, verify that the parsed text contains complete handwritten or printed content, and confirm OCR recognition accuracy meets expectations.
- Upload a single 100 MB case file, verify that parsing time is lower than the
PARSE_FILE_TIMEOUT_SECONDSconfiguration value, with no timeout error prompt. - Set
chunkSizeto 800 characters, split a petition notice containing titles and paragraphs, verify that chunking results retain the association between titles and corresponding paragraphs. - Test uploading an Excel-format petition statistical table, verify that the workflow node can normally read the field content in the table.
The values provided here are common starting points and should be measured 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 2026-09-28.