What the Data for This Category Looks Like
Daily documents for this category include growth experiment plans, user segmentation tables, outreach copy, and retention analysis reports. Data sources are user behavior logs, push systems, experiment platforms, and CRM systems. Most documents are updated daily or weekly. User segmentation tables contain fields such as user identifiers, outreach records, and behavior tags. Outreach copy marks push channels and trigger conditions. Retention analysis reports include periodic behavior statistics. Overall, documents use both structured table and long text formats.
Constraints on Document Parsing and Chunking
Documents in this category have both structured table and long text formats, and most are updated daily or weekly. This creates multiple constraints for the document parsing and chunking process. First, structured user segmentation tables require retention of the association between fields and rows. Avoid breaking data logic after chunking. Second, long text growth experiment plans need proper truncation. Avoid context breaks after chunking. Third, frequently updated documents require consistent chunking rules. Avoid differences in chunking across similar documents that disrupt subsequent retrieval and analysis.
Configuration Settings
These configurations apply to FastGPT v4.17.0 and later versions.
| Configuration Item | Recommended Value/Method | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Ensures semantic integrity for long texts such as growth experiment plans and retention analysis reports, avoiding context breaks after chunking |
chunkSplitter | Prioritize paragraphs | Adapts to documents with mixed structured and unstructured content such as user segmentation tables and outreach copy, preserving logical associations between content |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports uploading large batches of user behavior log files in a single operation, matching daily bulk import requirements |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Addresses parsing time for large user segmentation tables or multi-page experiment plans, avoiding timeout interruptions |
chunkSettingMode | Semi-automatic mode | Allows adjustment of chunking rules for different types of operation documents, balancing processing needs for structured tables and long texts |
Three Common Configuration Errors
- Phenomenon: After uploading operation documents larger than 10 MB, chunk vectorization fails for some chunks. Retrying multiple times partially resolves the issue. Cause:
UPLOAD_FILE_MAX_SIZEand chunking parameters are not configured correctly. Chunk boundaries for large files do not align with semantic logic, leading to errors during vectorization. - Phenomenon: When attempting to import online spreadsheet content, field associations in the table are not preserved correctly. Cause: Structured table parsing related configurations are not enabled. The default chunking logic breaks row and column associations in tables.
- Phenomenon: Exported dataset.csv files cannot be used directly to create new knowledge bases. Cause: Export files in question-and-answer format are mistakenly treated as structured data templates. The correct export format for operation documents is not selected.
How to Verify Successful Configuration
- Check that server environment variables
AIPROXY_API_ENDPOINT(root address of the AI Proxy service, for example http://aiproxy:3000; do not append paths such as /v1) andAIPROXY_API_TOKEN(AI Proxy administrator token, consistent with AI Proxy server configuration and not empty) are configured, and that parameter values meet official requirements. - Upload a test user segmentation table or retention analysis report. Verify that chunked content retains logical associations between fields or paragraphs.
- Upload a test document of approximately 10 MB. Confirm that parsing proceeds normally with no timeouts or error messages.
- View the chunk list in the knowledge base. Confirm that the length of each chunk falls within the preset
chunkSizerange.
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 2026-10-19.