What the Data for This Category Looks Like
The data sources for coking coal research reports mainly cover publicly disclosed research content from coal industry self-regulatory organizations, commodity trading platforms, professional consulting institutions, and upstream and downstream enterprises in the industrial chain. Document formats vary: some are plain text analysis reports, while others include structured data tables and visual charts. Each single piece of content includes modules such as core indicator summaries, supply and demand situation analysis, price trend judgments, and policy impact analysis. Update cycles are layered according to industry conventions, including weekly tracking, monthly summaries, and quarterly in-depth analysis content. The content includes coking coal-specific quality and circulation indicators, paired with industry-standard units of measurement.
Constraints Imposed on Source Citation and Traceability
The multi-source data and format differences of coking coal research reports require the traceability link to support multiple parsing rules and accurately match cited content to the specific location of the original source. Coking coal has unique indicators, so the traceability link must accurately associate the original disclosure content of corresponding fields to avoid confusion with data from other coal categories. There are layered update cycles for research reports: weekly tracking content and quarterly in-depth analysis have different timeliness requirements. The traceability link must simultaneously mark the release time and update cycle of the content to ensure that the timeliness of citations can be verified.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
ragRecallTopK | Top 8-12 results | Coking coal research reports have many indicator dimensions, so a sufficient number of recalled contents are needed to cover core analysis directions and avoid missing key traceability basis |
rerankTopN | Top 3-5 results | The effective associated content of coking coal research reports has a high concentration. Too many reranked results will increase the computational cost and complexity of traceability matching |
ragSimilarityThreshold | 0.75-0.85 | The semantic difference of coking coal-specific indicators is relatively high. This threshold can filter low-match irrelevant coal category research report content |
chunkOverlap | 150-250 characters | Core indicators of coking coal research reports are often distributed across sections. Overlapping chunks can retain complete contextual associations and improve traceability accuracy |
enableSourceCitation | Enabled | Source citation and traceability is a core functional requirement for coking coal research reports, so the specific location and release information of the original source must be forcibly displayed |
The parameter values provided on this page are conventional recommendations used as a starting point for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require specific analysis. It is recommended to test on your own samples before finalizing settings.
Three Common Misconfigurations
- Phenomenon: The number of recalled cited documents far exceeds the preset value, and a large number of non-coking coal category research report contents are included. Cause: The
ragSimilarityThresholdparameter is set too low, failing to filter low semantic matching irrelevant content, resulting in an overly large traceability scope. - Phenomenon: The cited content in the response is not the most comprehensively relevant coking coal research report content in the knowledge base. Cause: The reranking step is not enabled, or the
rerankTopNparameter is set too small, failing to include the most relevant traceability candidates in the final display range. - Phenomenon: The citation description generated by the large model does not match the specific location of the original document, with incorrect chapters, tables, or indicator descriptions. Cause: The
chunkOverlapparameter is set unreasonably, with insufficient chunk overlap rate, causing core indicators and analysis context to be split, making it impossible to establish accurate traceability associations.
How to Verify Proper Configuration
- Upload 1 to 2 publicly available coking coal research reports and other coal category research reports, perform retrieval tests, check the proportion of coking coal-related content in the recall results, and adjust corresponding parameters to the required range.
- View the source content display module of retrieval results, check whether each citation result marks the release time, source subject, and specific chapter location of the original document, confirming that the traceability configuration has taken effect.
- Generate a question related to core coking coal indicators, check whether the cited content in the response matches the segmented content of the original research report, and adjust chunk-related parameters to optimize association accuracy.
- Test the recall effect of different retrieval thresholds, check the relevance and quantity of cited content, and ensure that the configuration meets the business traceability accuracy requirements.
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.