Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
Data of this category mainly comes from archived ticket records of state-owned enterprises and government hotlines, standardized response template libraries, and publicly shared documents from internal knowledge bases such as Feishu and Yuque. Data sources include multiple formats such as structured ticket tables, standardized script text, and ticket screenshots with embedded images. Ticket records include fields such as ticket number, call time, problem category, user request, and closed response template. Response template documents are categorized by business scenario, with each entry linked to problem keywords and standard scripts. The data update frequency is adjusted based on hotline business volume, with incremental updates hourly or daily on a regular basis. Template documents are synced weekly or monthly.
What constraints do these characteristics impose on the "document parsing and chunking" link
Diverse data types are involved, including structured tables, image-containing documents and plain text. This requires the parsing link to support multi-format adaptation and image OCR extraction. Response templates are stored by business category, so chunking must retain classification metadata to avoid cross-category chunk confusion that affects recommendation accuracy. Ticket text varies widely in length: short requests are only tens of characters, while long requests can reach thousands of characters. Chunking parameters need to balance context completeness and retrieval efficiency. Some data sources such as Yuque public links require remote pulling, so the parsing timeout threshold needs to accommodate the processing time of long documents or batch files.
How to set the configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapt to the text length of hotline tickets and response templates, avoiding too fragmented chunks that lose context or overly long chunks that cause LLM context overflow |
chunkSplitter | Semantic segmenting | Adapt to the structured classification of response templates and the natural user requests of tickets, retaining semantic integrity |
customPdfParse | Enable image OCR parsing | Process screenshot documents in ticket attachments, extract text content from embedded images |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Accommodate batch imported ticket Excel and multi-dimensional documents, avoiding large file upload failures |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Accommodate the time requirements of long document or multi-file batch parsing, preventing premature timeout interruptions |
chunkSettingMode | Manual configuration | Adjust chunking parameters for different data sources (such as Excel tickets, Yuque links) to meet the needs of specific scenarios |
Three common errors
- Phenomenon: After parsing ticket documents with embedded images, only the text framework is retained, and image content is not extracted. Cause: The OCR parsing configuration of
customPdfParseis not enabled, and the image content extraction function is not activated. - Phenomenon: After importing a Yuque public share link, the system prompts that the data source cannot be parsed. Cause: The link is not publicly accessible, or the link includes login-required permission restrictions, which do not meet the data source requirements for direct parsing.
- Phenomenon: After batch importing Feishu multi-dimensional documents or Excel tickets, structured fields (such as problem category) are not retained, and business tags are lost in chunking results. Cause: The metadata retention configuration is not enabled, or when
chunkSettingModeis set to automatic adaptation, no field mapping rules are associated.
How to confirm the configuration is correct
- Upload a ticket document with embedded images. Verify if the parsed result includes text content extracted from images.
- Import a publicly shared Yuque knowledge base link. Confirm that the system does not prompt parsing failure and can normally pull document content.
- Set
chunkSizeto 1000 characters, upload a ticket text of 1500 characters. Verify if the chunking result is split into segments matching the set length. - View the parsing task logs. Confirm that no timeout errors occur during large file parsing, which complies with the setting of
PARSE_FILE_TIMEOUT_SECONDS.
The values provided 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.