What this category of data looks like
This category of data primarily comes from national government procurement websites, local government procurement websites, fiscal integration systems, purchaser 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 responses. Data fields include project identifiers, procurement content, qualification requirements, scoring rules, and more. The structure of different documents varies significantly: for example, bidding documents include general clauses and special clauses, while winning bid announcements include information about the winning supplier and transaction amount.
What constraints these characteristics impose on the "document parsing and chunking" link
Data for this category comes from multiple sources, including public web pages, system-exported files, and uploaded documents, with incremental content updated daily. This requires parsing tools to support multi-format input. Document structures vary greatly; some bidding documents include embedded images, tables, and fixed-format headers and footers. Chunking must preserve the integrity of key information. The priority of fields differs across documents: for example, qualification requirements in procurement requirements must be fully extracted first. If parsing omits information or chunks break, subsequent Q&A may fail to accurately match policy basis, affecting the compliance of the procurement process.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Government procurement documents often contain long paragraphs of policy clauses and procurement requirements. This range ensures single chunks have complete semantic meaning and avoid splitting critical rules. |
chunkSplitter | Split by paragraph + title | Government procurement documents have clear section titles and clause hierarchies. Splitting by title preserves document structure, making it easier to match specific clauses for subsequent Q&A. |
customPdfParse | Enable embedded image extraction | Bidding documents often include embedded qualification certificates, formatted attachments, and other images. Enabling this setting converts image content into text annotations for large model use. |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Some large-scale procurement projects have large bidding documents or bid evaluation reports. This value covers conventional upload requirements. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Complex PDFs or multi-page documents take longer to parse. 600 seconds avoids interrupting the parsing process due to timeout. |
Reference only; verify against the target FastGPT release and deployment documentation.
Three common mistakes
- Phenomenon: Parsed document chunks do not include embedded image information, so large models cannot access image content. Cause: The
customPdfParseconfiguration item is not enabled, or image extraction rules are not configured. - Phenomenon: Importing a public web link from a non-government procurement official channel results in a parsing failure prompt for unsupported data source type. Cause: The data source is not confirmed as a publicly accessible government procurement web page; links from non-official channels cannot be correctly identified.
- Phenomenon: Uploaded PDF files cannot be parsed, with a prompt indicating the file is too large or parsing timed out. Cause: The
UPLOAD_FILE_MAX_SIZEorPARSE_FILE_TIMEOUT_SECONDSparameters are not adjusted to values matching the document size and parsing duration.
How to confirm the configuration is correct
- Upload a typical government procurement bidding document, and check if the parsed text includes complete section titles and clause content.
- After enabling the
customPdfParseconfiguration, upload a document containing embedded images, and confirm that the parsing result includes text annotations or extracted content corresponding to the images. - Test importing a public web link from the government procurement website, and confirm that the parsing process completes normally without error prompts.
- Adjust the
chunkSizeparameter, then compare chunking results across different values, and confirm that the semantic completeness of single chunks meets expectations.
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.