What this category of data looks like
Data sources include official quote data from the Shanghai Gold Exchange and London Bullion Market Association, industry supply and demand analysis reports, central bank gold reserve announcements, and geopolitical analysis documents. Update frequencies vary significantly: spot real-time quotes update every 15 minutes, weekly supply and demand reports release on a fixed schedule, and central bank reserve data updates monthly. Documents contain structured product quote tables, semi-structured market analysis sections, and unstructured policy interpretation content. Fields include product code, trading unit, latest transaction price, position volume, and more. Common units are yuan/gram, US dollars/ounce, kilogram, and ton.
Constraints for Document Parsing and Chunking
High-frequency, multi-source data requires the parsing process to accurately match structured fields, preventing misalignment between quote data, units, and product codes. Coexisting multi-format documents require the parsing module to support fixed-format reports, freely formatted research reports, and other document types. It must adapt to different table extraction rules. Diverse units require retaining unit associations after parsing, avoiding confusion between yuan/gram and US dollars/ounce. The chunking process splits real-time timestamped data into independent chunks by release time, ensuring data timeliness. The chunking process splits semi-structured research report sections by theme instead of fixed-length splitting, maintaining logical integrity.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
PARSE_TABLE_STRICT_MODE | Enabled | Precious metals documents have many structured table fields; strict mode prevents misalignment between quote data, units, and product codes |
CHUNK_SIZE | 800–1200 characters | Adapts to the typical paragraph length of precious metals research reports, balancing context completeness and retrieval accuracy |
PARSE_UNIT_AWARE | Enabled | Covers multi-unit scenarios such as yuan/gram and US dollars/ounce, automatically associating numerical values with their corresponding units |
PARSE_TIMEOUT_SECONDS | 300 seconds | Meets parsing duration requirements for large research reports and batch historical quote documents, avoiding timeout interruptions |
ENABLE_CHUNK_DUPLICATE_CHECK | Disabled | Retains the original order of custom chunks, preventing index disorder caused by the system automatically deleting duplicate chunks |
CHUNK_OVERLAP | 50–100 characters | Maintains context continuity across chunks, adapting to logical connections in price trend analysis paragraphs |
The parameter values provided on this page are common recommendations used as a starting point for configuration. Actual values are affected by material format, data volume, and business rules. Specific issues require individual analysis, and it is recommended to test on your own samples before finalizing settings.
Three Common Configuration Errors
- Symptom: After custom chunking, the knowledge base automatically deletes duplicate chunks. This causes the index order to mismatch the custom chunking results. Cause: The
ENABLE_CHUNK_DUPLICATE_CHECKconfiguration item is not disabled. The system enables duplicate chunk filtering by default. - Symptom: Precious metals quote tables parse into scattered text. Retrieval results cannot be presented in table form. Cause:
PARSE_TABLE_STRICT_MODEis not enabled. Relaxed mode fails to recognize the fixed-format table structure of precious metals documents. - Symptom: After calling the knowledge base creation API, status information for parsing in progress, ready, or failed cannot be obtained. Cause: The status return switch is not configured in the API request parameters, or the corresponding status callback interface is not connected.
How to Verify Correct Configuration
- Upload a standard Shanghai Gold Exchange quote PDF document. Check if parsed table fields fully match product, price, and unit information from the original document.
- Upload a custom-chunked research report document. Verify that chunk order in the knowledge base matches custom chunking results, and confirm no chunks are automatically deleted.
- Call the document parsing API. Check if the returned
parse_statusfield includes status identifiers for parsing in progress, ready, and failed. Confirm normal status synchronization. - Upload a document containing multiple units. Check if parsed chunks retain associations between numerical values and units, with no unit misalignment.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.