What the data for this category looks like
This category of data primarily comes from state-owned assets supervision information systems, enterprise asset management systems, property rights exchange systems, valuation firm reports and financial statements. Commonly processed documents include property registration forms, asset valuation reports, property rights transfer plans, on-site transaction announcements, and asset disposal approval documents. Most documents contain structured fields such as asset numbers, affiliated units, and valuation values. Some documents include multi-column tables and embedded images. Data is updated monthly, and some temporary transaction documents are updated on demand.
What constraints these characteristics impose on the document parsing and chunking step
Property-related documents contain a large number of structured fields. When chunking, the association between fields and their corresponding context must be retained to avoid splitting that prevents matching key information. A high proportion of documents have multi-column tables and embedded images. Conventional chunking logic may destroy table structures or lose the association between images and main text. Monthly batch-updated asset reports and temporary transaction documents coexist. Parsing efficiency for bulk tasks and parsing duration for individual complex documents must be balanced to prevent timeouts or incompatible chunking logic.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the field length and multi-paragraph structure of property-related documents, avoiding loss of key transaction and ownership information after splitting |
chunkSplitter | Split by paragraph + heading | Matches the heading hierarchy structure of asset valuation reports and property rights transfer plans, retaining logical coherence of documents |
customPdfParse | Enable image recognition mode | Adapts to embedded asset photos and attached images in valuation reports, meeting the image-text association requirements of inquiry and response scenarios |
UPLOAD_FILE_MAX_SIZE | 500 MB | Compatible with large-scale batch-imported property transaction announcements and annual asset reports, preventing upload failures |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the parsing time required for complex asset valuation reports, preventing parsing interruptions due to timeouts |
Three common mistakes
- Symptom: Images are not output with associated main text after parsing, or image fields are empty. Cause: The image recognition mode of
customPdfParseis not enabled, or the chunking logic for image-text association is not configured. - Symptom: Column order is disorganized and fields are misaligned after parsing multi-column tables. Cause: The splitting rules of
chunkSplitterare not adjusted for multi-column documents, and the default splitting logic cannot adapt to the table's column structure. - Symptom: Parsing times out when uploading large asset reports in batches. Cause: The value set for
PARSE_FILE_TIMEOUT_SECONDSis lower than the actual parsing time required for the documents, and insufficient processing time is reserved.
How to confirm the configurations are set correctly
- Upload a single asset valuation report that includes embedded images, and check if the parsing result contains content associated with images and main text.
- Upload a property rights transfer announcement with a multi-column table, and verify that the column order of the parsed table matches the original document.
- Upload an asset report that is close to the
UPLOAD_FILE_MAX_SIZEthreshold, and confirm that the upload and parsing processes complete normally. - Check the parsing logs to confirm that no timeout errors triggered by the
PARSE_FILE_TIMEOUT_SECONDSsetting occur.
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.