What data for this category looks like
Daily data for level content creation teams comes from level editors, art resource libraries, player test data, and defect systems. Daily documents include level design drafts, resource requirement sheets, test feedback forms, and version submission records. Most document structures include hierarchical chapters, uniquely identified level IDs, resource paths, and test level fields. Units include characters, MB, level tags, and similar. Data update frequency is weekly or per version. Field differences between different documents are significant. Structured documents have clear column items, while unstructured documents are mostly paragraph-style descriptions.
What constraints do these characteristics impose on document parsing and chunking?
The multi-source, multi-format data characteristics of this category require document parsing to support multiple file types including PDF, Excel, CSV, and text. High-frequency weekly or per-version updates require parsing processes to have batch processing capabilities, to avoid delaying research and development progress due to update delays. Documents contain key associated fields such as level IDs and resource paths. Chunking operations must retain contextual associations between fields, to avoid splitting that breaks the integrity of level logic. The large format differences between structured resource requirement sheets and unstructured test feedback forms require adaptation to different chunking rules, otherwise critical information may be lost.
How to configure settings
The following configurations are formulated based on the parameter rules of FastGPT v4.17.0:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the information density of single paragraphs in level design drafts, avoids splitting that breaks complete context of level paths and resource associations |
chunkSplitter | Split by title + paragraph | Matches the hierarchical structure of level documents, retaining the associative relationship between chapters in design drafts |
customPdfParse | Enabled | Adapts to imported level design PDF documents, retains the original document layout and field structure |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Compatibility with large art resource-related documents and version submission record packages |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the time required for batch parsing of multiple files, avoids parsing timeouts for large documents |
chunkSettingMode | Manual mode | Allows adjustment of chunking parameters for different types of level documents, such as test feedback forms and resource requirement sheets |
Three common configuration errors
- Phenomenon: Enabling
customPdfParsewhen importing PPT or DOC documents leads to abnormal parsing results. Reason:customPdfParseonly applies to PDF format documents, and does not adapt to parsing rules for other office documents. - Phenomenon: The knowledge base cannot import Excel files, and CSV only supports two columns of data. Reason: Corresponding custom parsing configuration for the format is not enabled, and chunking adjustments for multi-column structured data are not made.
- Phenomenon: Knowledge base Q&A returns content that deviates from preset question-answer pairs. Reason:
chunkSettingModeorchunkSizeis not set correctly, leading to incorrect splitting of the context of question-answer pairs.
How to confirm configurations are correct
- Upload a single level design PDF document, check if the parsed chunks retain the association between chapter titles and corresponding paragraphs.
- Upload an Excel-format resource requirement sheet, check if the parsed result completely extracts multi-column field information.
- Upload a batch of version submission records, wait for parsing to complete and confirm there are no timeout errors in task status.
- Import preset question-answer pair documents, initiate test queries, and confirm returned content matches preset answers.
The values provided here are common starting points, and should be evaluated against samples relevant to the specific deployment.
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.