What the data for this category looks like
State-owned capital investment companies’ business process-related data comes from multiple sources: state-owned assets supervision and reporting systems, internal project approval process documents, Feishu collaborative spreadsheets, network segment management records, and similar resources. Data updates follow two rhythms: fixed-cycle reporting (for example, monthly supervision reports) and temporary project documents (such as special approval files). Most documents use multi-column structured table layouts. They include fields like project number, approval node, responsible person, approval time limit, IP network segment, and subnet mask, paired with long-text approval opinions. Field types include numbers, dates, and text. Units include days, count, and standard IP segment format, among others.
What constraints these characteristics impose on document parsing and chunking
The default parsing logic for multi-column structured tables only recognizes the first and last columns. This loses intermediate fields such as approval node and IP network segment information. Long-text approval opinions have wide length ranges. Too short chunks break the coherence of approval logic. Too long chunks reduce retrieval accuracy. For continuous numeric IP network segment fields, splitting must retain field relevance. Do not split a single network segment into multiple chunks. When Feishu online spreadsheets act as data sources, the system must support online pulling and full parsing of all columns. Scenarios mixing fixed-cycle and temporary documents need to adapt to parsing requirements for documents with different update rhythms.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the average length of approval opinions and project descriptions in business documents of state-owned capital investment companies, preventing overly fragmented or overly long single chunks from impacting retrieval |
chunkSplitter | Split by paragraph + header | Adapts to multi-column structured tables and long-text approval opinions, retains field relevance, and avoids splitting that disrupts the logical structure of network segment record tables and approval processes |
customPdfParse | Enable multi-column parsing mode | Adapts to the multi-column layout of state-owned assets supervision reports, solves the problem that the default parsing only identifies two columns, and fully extracts all fields |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Covers the parsing time of large project approval documents and batch network segment record tables, preventing timeout failures |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapts to the total size of batch-imported state-owned assets supervision report packages and multiple network segment record tables, supporting batch document uploads |
chunkSettingMode | Manual configuration | Sets chunking parameters separately for different document types (such as table-type and long-text documents) to adapt to mixed document scenarios |
Three common errors
- After uploading a multi-column Excel file, the knowledge base only extracts the first two columns of content. Cause: The multi-column parsing configuration for
customPdfParseis not enabled. The default parsing logic only identifies the first and last columns. - After uploading a network segment record table file, complete network segment information cannot be matched during retrieval. Cause: The continuous IP segment field was split during chunking. This destroys field relevance, preventing complete network segment data from being associated during retrieval.
- When batch importing large approval documents, the parsing task shows a timeout failure. Cause: The
PARSE_FILE_TIMEOUT_SECONDSparameter was not adjusted. The default timeout duration is insufficient to cover the parsing time of large documents.
How to confirm the configuration is correct
- Upload a test Excel file with more than 3 columns. Verify that the number of fields in the parsed knowledge base matches the original table.
- Upload a single approval opinion document with a length exceeding 500 characters. Check that the chunking result length falls within the preset
chunkSizerange. - Submit a batch large document parsing task. Confirm that no timeout errors appear in the task status.
- Import a Feishu online spreadsheet data source. Check that all columns of data have been successfully imported and stored.
The values provided here are common starting points and should be evaluated 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.