What This Category's Data Looks Like
Data sources for coke financing daily reports include the Coke Branch of China Coal Industry Association, public monitoring data from domestic coastal coke ports, and procurement filing information from major steel mills. The update cadence is full data for the previous trading day, released every early morning. The document is a single-page structured report with four modules: regional spot market trends, port inventories, steel mill procurement, and railway shipments. Fields include the daily average spot price, total port inventory, major steel mill procurement volume, and regional price difference, with units of yuan/ton, ten thousand tons, tons, and yuan/ton respectively.
Constraints for Knowledge Base Retrieval and Recall
Multiple data sources lead to format differences in coke data across channels. Format alignment of cross-source data must be completed during the recall stage to avoid confusion between field names and units. The daily full update cadence requires the knowledge base incremental update configuration to match the update cycle, to avoid repeatedly loading already synchronized historical data. The structured report structure requires the retrieval system to support field-level precise matching. Using full-text fuzzy matching will fail to accurately locate target market data. Entities such as steel mills and ports associated with coke downstream must have their retrieval weights configured, to prevent irrelevant coal category data from being recalled.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
RECALL_TOP_K | Top 10 results | Coke financing daily reports have many fields per document. Too many recall results will exceed the conversation context window and impair model understanding |
PARSE_STRUCTURED_TABLE | Enabled | Daily reports use structured report formats. Enabling this setting extracts field-level retrieval capabilities and improves precise matching efficiency |
UPLOAD_FILE_MAX_SIZE | 500 MB | Adapts to scenarios where multiple coke daily reports are uploaded in batches, avoiding exceeding the system's default upload limit for a single request |
SIMILARITY_THRESHOLD | 0.75–0.85 | Precise matching of field names and units is required. This range filters low-correlation non-target data |
INCREMENTAL_UPDATE_INTERVAL | 24 hours | Matches the daily update cadence of the daily reports, avoiding repeated loading of already synchronized historical data |
RE_RANK_TOP_N | Top 3 results | Re-ranking for structured data should focus on core market fields, reducing interference from redundant results for the model |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require targeted analysis. It is recommended to test on your own samples before finalizing settings.
Three Common Mistakes
- Symptom: Retrieval of coke financing daily report content returns empty in conversations, but the knowledge base backend test returns results normally. Cause: The precise matching rule for
SIMILARITY_THRESHOLDis not configured, or conversation context interferes with semantic matching of retrieval keywords. - Symptom: After uploading multiple coke daily reports, some documents cannot be retrieved. Cause: The
PARSE_STRUCTURED_TABLEconfiguration is not enabled. Structured reports are not correctly extracted for fields, causing keywords to fail to match target content. - Symptom: A
413 Request Entity Too Largeerror occurs when uploading daily reports in batches. Cause: TheUPLOAD_FILE_MAX_SIZEconfiguration is not adjusted, exceeding the system's default file upload limit.
How to Verify Successful Configuration
- Upload a single coke daily report, use the knowledge base backend to test retrieval for "daily average spot price", confirm that content corresponding to the field is returned.
- Adjust the value of
SIMILARITY_THRESHOLD, confirm that the precision of retrieval results meets business requirements through multiple rounds of testing. - Upload 3 or more coke daily reports in batches, check the file parsing status in the backend, confirm that there are no failed entries.
- Initiate a conversation test, input retrieval terms including "port inventory" and "steel mill procurement volume", confirm that returned results include content corresponding to the fields.
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.