What the Data for This Category Looks Like
Data related to service applications for basic telecommunications operations is primarily sourced from official business manuals, tariff package documents, process guidance PDFs, and temporary policy notices. Updates are conducted regularly alongside package adjustments, process optimizations, or new regulation releases. Each update covers all documents for one or multiple service types. Document structures include structured tables (such as tariff tiers and eligibility checklists), long-text process descriptions, and image-text guides with operation steps. Fields covered include package tiers, effective periods, charging standards, application channels, and more. Units are mostly yuan/month, GB, minutes, workdays, and similar units.
Constraints Imposed by These Characteristics on the Document Parsing and Chunking Stage
The mixed structure of structured tables and long text in service documents requires that table field associations are not broken during chunking. Failure to preserve these associations will prevent users from matching complete eligibility requirements during retrieval. Frequently updated documents require frequent re-parsing, while large full business manuals create resource pressure during upload and parsing. A large number of repeated rule prefixes (such as "To apply for this service, the following requirements must be met") exist in documents. If context is not retained during chunking, individual text chunks cannot be understood independently. Additionally, some documents contain cross-page process descriptions, so chunking must avoid truncating complete process nodes.
How to Set Configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Basic telecommunications service documents mostly mix rule descriptions and tables; this length retains the context of a single package or complete process node |
chunkSplitter | Split by paragraph + heading | Business documents have clear process nodes and package headings; this split avoids chunking across nodes or packages |
customPdfParse | Enable structured table parsing | Basic telecommunications documents contain numerous tariff and eligibility tables; enabling this preserves complete field integrity within tables |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapt to upload requirements for full business manual documents, avoiding system restrictions triggered by large file uploads |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large business documents take longer to parse; this duration covers the complete parsing and upload process |
chunkSettingMode | Manual configuration | For business documents with mixed structures, manual configuration allows flexible adjustment of chunking rules to adapt to different types of business documents |
Three Common Mistakes
- Uploading large PDF files triggers an "offset out of range" error at 90% parsing progress. The cause is failure to adjust the
UPLOAD_FILE_MAX_SIZEandPARSE_FILE_TIMEOUT_SECONDSparameters, leading to large file parsing timeout or insufficient memory allocation. - A large volume of duplicated package tier content appears after chunking. The cause is failure to set a reasonable
chunkSplitterrule; fixed-character chunking truncates the context boundary of duplicated content, resulting in duplicate fragments in adjacent chunks. - Table fields are empty after document parsing, and complete eligibility requirements cannot be extracted. The cause is failure to enable the structured parsing function of
customPdfParse, with pure text chunking directly causing loss of structured field information within tables.
How to Verify Proper Configuration
- Upload a business PDF containing a tariff table. Confirm parsed knowledge base entries retain complete table row and column structures with no missing fields.
- Upload a single business manual document not exceeding 1000 MB. Confirm parsing progress completes normally to 100% with no error prompts.
- Randomly sample chunked text fragments. Confirm chunking does not split across process nodes or package tiers, and individual chunks can independently express complete service information.
- Test uploading multiple updated business documents. Confirm parsing time does not exceed 600 seconds with no timeout errors.
The values given 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.