What data looks like for this category
The data for this category mainly comes from government and enterprise service guidelines, policy normative documents, public inquiry work orders, and official reply documents. Data updates follow no fixed schedule, and are synchronized with policy adjustments, service process updates, or new inquiry cases. Documents fall into two categories: structured and unstructured. Structured documents mostly use chapter-based formatting, and include fields such as document numbers, accepting units, and processing time limits. Unstructured content consists of conversational inquiry replies, containing user questions and official responses. Some documents embed standardized tables and policy attachments.
What constraints these characteristics impose on document parsing and chunking
Parsing tools must accurately identify title hierarchies for structured service guidelines with chapter-based formatting, to avoid chunk truncation of policy clauses or process steps. Embedded standardized tables and attachment images require full extraction of metadata and visual content. Missing this data removes key details needed for service handling. For unstructured conversational documents, teams must distinguish between user questions and official replies, to prevent chunking from mixing interactive content. Some documents have long single-page lengths. Administrators must set a controlled timeout threshold for parsing and chunking, to avoid parsing failure caused by overly long content. Document fields such as document numbers and processing time limits must be retained alongside each chunk. This ensures complete service information can be linked after retrieval.
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 and enterprise service documents often contain long paragraphs of policy clauses and process descriptions. This range ensures chunks contain complete semantic units, and avoids splitting service steps. |
chunkSplitter | Split by title and paragraph boundaries | Adapts to the chapter-based formatting of structured service guidelines, ensures chunks match business logic boundaries, and avoids random text truncation. |
customPdfParse | Enable built-in OCR parsing mode (replace default parsing tool) | Government and enterprise documents often include scanned policy files and embedded images. This configuration replaces the built-in parsing tool and enables OCR mode to fully extract service information from images, solving the problem of images not displaying properly. |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapts to large-volume documents such as policy compilations and work order collections uploaded in bulk by government and enterprise users, and prevents upload failure due to exceeding file size limits. |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Government and enterprise documents are often combinations of long text and multiple attachments. This timeout range ensures complex documents complete parsing and chunking, and prevents mid-process interruptions. |
chunkSettingMode | Manual configuration | Adapts to the diverse formatting of government and enterprise documents, allowing adjustment of chunking rules for different document types such as structured guidelines and conversational work orders, to meet business scenario needs. |
Three common mistakes
- Embedded images cannot be viewed after uploading a PDF, and the image fields in the parsing result are empty. The cause is that the OCR parsing mode of
customPdfParseis not enabled, so only text content is extracted and image resources are not processed. - Search results only return text fragments, and do not associate corresponding metadata such as document numbers and processing time limits. The cause is that the rule to retain metadata alongside chunks is not configured during chunking, so only plain text content is extracted.
- An upload failure prompt is triggered when bulk uploading Excel documents. The cause is that
UPLOAD_FILE_MAX_SIZEis not adjusted to adapt to large-volume Excel tables, or parsing support for the corresponding format is not enabled.
How to confirm configurations are set correctly
- Upload a government and enterprise policy PDF containing embedded images, and check whether the parsing result includes image previews and corresponding text content.
- Upload a structured service guideline document, and check whether the chunking result uses chapter titles as boundaries, and does not truncate complete process steps.
- Upload a single large-volume document, confirm that the parsing process does not experience timeout interruptions, and that the number of chunks meets expectations.
- Test the search function, confirm that the returned chunk results include corresponding metadata such as document numbers and processing time limits.
The values provided here are common starting points, and should be evaluated against samples specific to the deployment context.
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.