What this category's data looks like
Daily documents for this category include territorial spatial master plans, detailed plans, special plans, planning implementation evaluation reports, public opinion summaries, and more. Data sources are the territorial spatial basic information platform, natural resource survey and monitoring data, population and economic statistical results, and remote sensing images. The data update cycle is annual. Most documents are multi-chapter long texts, with spatial coordinate tables and area statistical fields attached. Used units include hectares, square kilometers, and ten thousand people. Some documents embed remote sensing image thumbnails and vector data attachments.
Constraints on document parsing and chunking
Annual update documents have long lengths and embed multiple types of spatial data attachments and statistical tables. The chunking process must retain the contextual association between spatial fields and statistical units, to avoid losing the correspondence between spatial coordinates and indicators. Documents with mixed multi-source data have diverse formats. The system must adapt to structures such as PDF files, vector data attachments, and nested tables. It must also handle mixed arrangements of short texts like public opinion summaries and long reports, to prevent incorrect splitting of short texts across different chunks. Additionally, annual batch document processing requires controlling the parsing duration per document, to avoid process interruptions caused by timeouts.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 1000–1500 characters | Adapts to the contextual association needs of planning report long texts and spatial tables, avoids separation of spatial indicators and coordinates after splitting |
chunkSplitter | Split by chapter + paragraph | Matches the multi-chapter structure of planning documents, retains the integrity of planning division and indicator item paragraphs |
customPdfParse | Enable table recognition mode | Adapts to the large number of area statistics and spatial coordinate tables in planning reports, improves the accuracy of table content parsing |
UPLOAD_FILE_MAX_SIZE | 500 MB | Meets the upload requirements for single large planning reports and multi-attachment packaged documents |
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Meets the duration requirements for long text parsing and multi-attachment processing, avoids timeouts during batch document processing |
chunkSettingMode | Manual configuration | Adapts to the special structure of documents in this category, flexibly adjusts chunking rules to match the contextual requirements of spatial data |
Three Common Mistakes
- Symptom: The knowledge base chunk display works normally, but Markdown formatted content cannot be rendered in context references. Cause: The table recognition and Markdown output switches for
customPdfParseare not enabled, so only plain text content is retained in the parsed text. - Symptom: Different vector models are used locally and on the server, and index mismatch occurs after uploading chunked documents. Cause: The chunk configuration and the input length requirements of the vector model are not unified, so the chunk character count does not match the model's adaptation range.
- Symptom: After upgrading the version, passing a public link of a planning report fails to extract document content, and only empty results are returned. Cause: The link parsing whitelist configuration is not updated, or the verification rules for link formats in the new parsing tool have been adjusted, and the online storage link format of planning documents is not compatible.
How to Verify Correct Configuration
- Upload a typical planning implementation evaluation report, check whether the chunk list retains the contextual association between chapter titles and spatial tables, and confirm that spatial indicators are not split across different chunks.
- Upload a test document within 500 MB, verify that the upload and parsing process completes normally, and confirm that the
UPLOAD_FILE_MAX_SIZEconfiguration takes effect. - Pass a planning document with a public link, verify that the main text content and table information can be extracted after parsing, and confirm that the link parsing function works normally.
- Adjust
chunkSizeto 1200 characters, compare the text integrity before and after chunking, and confirm that the chunking rule matches the structural characteristics of documents in this category.
The values listed above 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.