What the data for this category includes
The data for this category comes from ticketing systems, chat logs, product logs, and customer contract SLAs, with a daily or real-time update frequency. Commonly processed documents include customer tickets, issue triage tables, processing records, and service reports. Tickets contain fields such as problem description, associated product module, customer feedback time, and triage tags. Triage tables include standardized entries such as risk level and response time requirements. Processing records associate ticket IDs and responsible person information.
What constraints these characteristics impose on document parsing and chunking
Dispersed data sources with real-time update support require the parsing module to adapt to batch access of multiple formats. The module must also support incremental synchronization workflows for document updates. Tickets and processing records have field association attributes. Chunking must retain contextual relevance to avoid losing field association information after splitting. Standardized entries in issue triage tables require precise chunking to support subsequent retrieval matching. Time limit clauses in customer contract SLAs must fully retain key content and not be overly split. Real-time updated chat logs and tickets produce mixed short and long text content. Chunking logic must adapt to documents of varying lengths to avoid redundancy or information loss.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the typical text length of tickets and processing records, retains problem descriptions and contextual associations |
chunkSplitter | Split by title and paragraph boundaries | Tickets and triage tables have structured titles. Splitting by boundaries avoids damaging field association |
customPdfParse | Enabled | Some customer tickets and service reports are submitted in PDF format. Enabling this allows complete parsing of text and embedded table content |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapts to the batch upload limit of FastGPT v4.17.0, supports large service reports and contract documents to avoid upload failures |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large PDF documents require longer parsing time, avoiding timeout interruptions of the parsing workflow |
chunkSettingMode | Manual configuration | Document structures for this category vary widely. Manual configuration allows precise matching of chunking requirements |
Three common configuration mistakes
- Phenomenon: After uploading Feishu multi-level directory documents, only the top-level directory content can be retrieved, and subdirectory content is missing. Cause: Recursive parsing configuration for nested directories is not enabled, so subdirectory documents are not included in the parsing scope.
- Phenomenon: Calling the API to upload a document shows a successful call, but no text content is returned in the parsing result. Cause: The
customPdfParseparameter is not configured correctly, or the correct file parsing trigger identifier is not included in the API request, resulting in the parsing workflow failing to start normally. - Phenomenon: A split error log is triggered when parsing large service reports. Cause: The
chunkSizeconfiguration value is too small, causing long text chunking to exceed the system splitting threshold, or thechunkSplitteris not configured to split by paragraph boundaries, resulting in conflicting splitting logic.
How to confirm configurations are set correctly
- Upload a test document containing multi-level directories. Check if the parsed chunking results include all subdirectory content to confirm that the recursive parsing configuration is effective.
- Call the API to upload a single test document. Compare the parsing results with manual uploads to confirm that the text content returned by the API parsing workflow is complete.
- Upload a document exceeding the preset
UPLOAD_FILE_MAX_SIZE. Verify whether the upload workflow triggers a reasonable intercept prompt to confirm that the size limit configuration is effective. - Adjust the
chunkSizeparameter to different ranges. Upload a long-text ticket document, check if the chunked content retains complete contextual associations to confirm that the chunking logic meets expectations.
The values provided are common starting points and should be evaluated against relevant sample sets for the deployment scenario.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.