What data for this category looks like
Thermal coal research report data mainly comes from supply and demand reports released by industry regulatory authorities, real-time market data from large coal trading platforms, and special analysis documents from securities research institutes. Updates primarily follow weekly and monthly report cycles, with temporary research reports issued alongside sudden policy or market changes. Document structures include standardized market tables (covering price, inventory, and transportation capacity data), policy interpretation paragraphs, supply and demand balance analysis, and price trend charts. Core fields include flat price, pithead price, and port inventory, with units mostly being yuan/ton, ten thousand tons, and ten thousand tons per day.
What constraints do these characteristics impose on the document parsing and chunking link
Thermal coal research reports have a high proportion of structured data, along with a large number of specialized fields and units. The parsing link must accurately identify the structure of tables and embedded charts to avoid data extraction errors. Long text analysis paragraphs and short data entries appear alternately. During chunking, balance must be maintained between contextual completeness and retrieval accuracy to prevent core market data from being split. Temporary research reports updated in real time require the parsing process to have high efficiency, to avoid harming retrieval timeliness due to parsing delays. In addition, research reports from different sources may have inconsistent field units, and unified adaptation must be completed during the parsing phase.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
PARSE_PDF_USE_MINERU | Enable via checkbox | Most thermal coal research reports contain complex nested tables and embedded charts. MinerU preserves complete text hierarchy and table structure |
chunk_size | 800–1200 characters | Research reports include both long logical analysis sections and short data entries. This range balances contextual completeness and retrieval accuracy |
chunk_overlap | 100–150 characters | Prevents critical information from breaking across chunks, and adapts to coherent logical paragraphs and associated data in research reports |
TABLE_VECTOR_ENABLE | Enable via checkbox | Core decision-making data in thermal coal research reports is mostly in table form. Enabling this feature converts tables into structured vectors to improve retrieval precision |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Single thermal coal research reports are typically 10 to 30 pages long. This duration covers the full parsing and chunking process |
UPLOAD_FILE_MAX_SIZE | 50 MB | Adapts to the scale of single research report collections or batch uploaded documents, and avoids parsing timeouts or excessive resource usage |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material format, data volume, and business rules. Specific issues require case-by-case analysis. Testing on relevant samples is recommended before finalizing settings.
Three Common Mistakes
- Symptom: Parsing fails after enabling
PARSE_PDF_USE_MINERU, returning504 Gateway Timeout. Cause: The locally deployed MinerU port is not correctly mapped to the FastGPT container network, or the parsing API address configured in FastGPT does not match the actual running port. - Symptom: Parsed research report table fields are empty, or units are not unified. Cause:
TABLE_VECTOR_ENABLEis not enabled, or MinerU specialized field extraction rules are not configured to adapt to exclusive fields of thermal coal such as flat price and port inventory. - Symptom: Retrieval results for core market data are scattered. Cause:
chunk_sizeis set too small, orchunk_overlapvalue is insufficient, causing coherent tables and analysis text to be split into multiple independent chunks.
How to Verify Correct Configuration
- Navigate to the FastGPT knowledge base settings page, confirm that
PARSE_PDF_USE_MINERUis enabled, and that the parsing API address matches the locally deployed MinerU running port. - Upload a standard thermal coal research report PDF, check the parsed text preview, and confirm that the table structure is complete, and core fields such as flat price and port inventory along with their corresponding units are correctly extracted.
- View the chunked result list, confirm that the length of each chunk falls within the configured range of 800–1200 characters, and that table chunks are not split into scattered text.
- Retrieve specific price data in the research report, confirm that chunks containing this data can be accurately recalled, and verify the effect of
TABLE_VECTOR_ENABLE.
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.