What this category of data looks like
Data sources include ledgers from domestic coking coal major producing region trading enterprises, daily delivery records from northern ports, and publicly available financing information from industry news platforms. Updates are completed at a fixed time each day. Documents primarily use structured table formats, and include fields such as date, main producing area, pit mouth quotation, port warehouse clearance price, futures contract settlement price, financing trading volume, financing cost range, and credit limit. Quotation units are yuan/ton, trading volume units are 10,000 tons, and credit limit units are 10,000 yuan. Each document corresponds to full daily market coking coal financing structured records, with no redundant unstructured text.
Constraints on the Knowledge Base Retrieval and Recall Link
The structured-dominant document structure requires the retrieval link to support both semantic matching and field-level filtering, avoiding loss of field information when relying solely on semantic recall. The fixed daily update rhythm requires configuring incremental sync tasks for the knowledge base to match the data update cycle, preventing delayed retrieval results. The design with multiple fields and clear units requires that recalled results fully retain the correspondence between fields and units, preventing unit mixing across different data categories. The format where each document corresponds to full daily records requires that retrieval can quickly locate target ranges using dimensions such as date and producing area. It also requires handling deduplication logic for multi-source same-field data to ensure result accuracy.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
Recall Count | Top 10-15 | Coking coal financing daily report records are compact in information. Too many recalled entries will cause context overload. 10-15 entries can cover mainstream retrieval needs |
Similarity Threshold | 0.72-0.85 | Structured data has high semantic matching accuracy. A threshold that is too low will introduce irrelevant records, while a threshold that is too high will miss valid matching items |
maxContext | 4096-8192 tokens | Sufficient structured field information must be retained to avoid losing key business data due to context truncation |
Chunk Length | 800-1200 characters | Single structured record has moderate length. Too long chunks destroy field association, while too short chunks split field information |
Incremental Sync Cycle | 1 time per day, executed at 7 AM | Matches the fixed daily update rhythm of coking coal financing daily reports, ensuring the timeliness of retrieval results |
Field Filter Switch | Enabled | Coking coal financing daily reports include multi-dimensional business fields. Enabling this switch supports precise screening by conditions such as producing area and price range |
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, and it is recommended to test on your own samples before finalizing the settings.
Three Common Configuration Mistakes
- Phenomenon: No added ollama qwen2.5 models appear in the text understanding model dropdown when creating a knowledge base. Cause: The model was not correctly mounted to the knowledge base's text understanding model pool, or the model loading format does not comply with FastGPT access specifications.
- Phenomenon: An error message "invalid configuration parameter name 'hnsw.iter'" is returned during knowledge base search. Cause: A non-existent parameter name was entered in the vector database configuration, triggering a configuration verification failure.
- Phenomenon: Retrieval results include coking coal financing daily report data from other enterprises. Cause: The knowledge base's permission isolation configuration was not enabled, or independent access permission control rules were not set for different enterprise knowledge bases.
How to Verify Successful Configuration
- Run an incremental sync task, and check whether the knowledge base update time matches the coking coal financing daily report update time.
- Submit a retrieval request that includes field conditions, and confirm that returned results only include records that meet the screening criteria.
- Trigger a knowledge base search, and check whether the similarity scores of returned results fall within the preset threshold interval.
- Verify permission control rules by using accounts with different permissions to submit retrieval requests, and confirm that only authorized knowledge base data can be accessed.
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.