What this category of data looks like
Data for this category originates from the China Government Procurement Network, local government procurement websites, fiscal integration systems, procurer budget systems, and supplier credit records, with a daily update frequency. Commonly processed documents include procurement requirements, procurement plans, bidding documents, tender documents, bid evaluation reports, winning bid announcements, procurement contracts, and question and answer responses. Four types of stakeholders are involved: procuring entities, fiscal department government procurement supervision offices, procurement agencies, and expert evaluators. Documents contain structured fields such as project number, qualification requirements, and bid evaluation criteria, as well as unstructured main text content.
What constraints do these characteristics impose on the document parsing and chunking step
Daily updated batch documents require parsing processes to support high-frequency batch processing and avoid timeout interruptions. Significant structural differences exist across multiple document types: bidding documents follow a chapter-based format, while question and answer responses follow a question-and-answer format, so splitting logic must adapt to different structures. Supplier credit records mix structured tables and unstructured text, so accurate distinction between fields and main content is required. Key information related to government procurement supervision is implicit across different documents, so chunking must retain contextual connections and avoid splitting core clauses. Single files may contain multiple related announcements, so cross-document incorrect splitting must be prevented.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Government procurement documents include both long paragraphs of bid evaluation criteria and short clauses of qualification requirements. This range balances contextual completeness and chunk granularity |
chunkSplitter | Split by headings and paragraphs | Government procurement documents often have clear chapter headings. Splitting by headings preserves business logic integrity |
customPdfParse | Enabled | Government procurement documents often contain complex formats such as tables, headers, and footers. Custom parsing preserves structured information |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large bidding documents or batch file parsing require longer processing time to avoid mid-process timeouts |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Batch tender documents or bid evaluation reports for government procurement projects may occupy significant storage space |
chunkSettingMode | Manual configuration | Significant structural differences exist across different government procurement documents. Manual configuration allows flexible adjustment of chunking rules |
Three common mistakes
- Phenomenon: PDF parsing tasks return an error with status code 500. Cause: The
customPdfParseconfiguration is not enabled, or thePARSE_FILE_TIMEOUT_SECONDSsetting is too short, causing large bidding documents to fail to complete parsing. - Phenomenon: Unable to specify PDF parsing parameters when calling the knowledge base file collection creation API. Cause: The
customPdfParseparameter is not included in the request body. - Phenomenon: Uploaded files cannot be directly passed to the large model node. Cause: The system default parsing process is not skipped, and custom parsing trigger rules are not configured.
How to confirm configurations are correct
- Upload a single government procurement bidding document and verify that the parsed chunks retain chapter headings and corresponding main content.
- Check the running logs of batch parsing tasks to confirm no timeout interruptions occur.
- Call the knowledge base file collection creation interface to verify that the
customPdfParseparameter can be passed and takes effect. - Upload a single supplier credit record file and verify the accuracy of splitting between structured tables and unstructured text.
The values provided 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.