What the Data for This Category Looks Like
Refractory material data related to finance, insurance, and wealth management industries mainly comes from quality inspection reports of manufacturing enterprises, product specifications, national and industry standard documents, and customer technical consultation records. Update cycles are set based on new product launches, standard revisions, or application scenario expansions. A single update includes multiple sets of physical and chemical parameters and application cases. Most documents have fixed fields: product grade, chemical composition (such as Al₂O₃ content), physical properties (such as bulk density, refractoriness), application scenarios, packaging and storage requirements. Physical performance parameters have clear units, such as g/cm³, ℃.
Constraints on Knowledge Base Retrieval and Recall From These Characteristics
Retrieval must precisely match parameter names and units for fixed fields and units, to avoid invalid recall results caused by unit conversion errors. Segmented retrieval must retain data relevance for multiple sets of related physical and chemical parameter paragraphs, to avoid incomplete presentation of product performance after splitting. The knowledge base must use incremental update mode to handle the feature of low update frequency but large single update volume, reducing resource consumption from full parsing. In marketing scenarios, retrieval must combine application scenario keywords. This requires retrieval rules to cover both product attribute text and application scenario text, to meet customer acquisition needs for finance, insurance, and wealth management industries.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Refractory material documents often contain multi-page nested physical and chemical data tables, with longer parsing time than general documents |
Segment Length | 800–1200 characters | Refractory material performance parameter paragraphs are long. Too short segment length will break the connection between chemical composition and application scenarios |
Number of Recalled Items | Top 6 | Meets customer acquisition needs for marketing content in finance, insurance, and wealth management industries. 6 results can cover different application scenarios |
Similarity Threshold | 0.72–0.78 | Professional parameter matching requires high accuracy. This range can filter irrelevant non-professional search results |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports uploading large industry standard PDFs or batch quality inspection reports, adapting to the document scale of refractory materials |
Reranked Returned Items | Top 3 | Marketing scenarios prioritize displaying core matching content, to avoid excessive results interfering with judgment |
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. Each situation should be analyzed individually. It is recommended to test on your own samples before finalizing.
Three Common Mistakes
- Phenomenon: After uploading images or PDFs containing images, knowledge base search results do not include text associated with images, and the interface shows parsing failed. Cause: The image OCR parsing switch in FastGPT was not enabled. Valid text from product appearance diagrams and kiln application schematic diagrams in refractory material documents was not extracted.
- Phenomenon: After exporting the full knowledge base and importing it to a new server, some custom fields (such as product grade prefix codes) are lost, and precise matching of target products fails during retrieval. Cause: Complete metadata fields were not selected during export, only plain text content was exported, and custom field information was not included.
- Phenomenon: After setting
PARSE_MAX_PARAGRAPH_DEPTHin version 4.9.10, nested table content in long documents is not fully parsed. Cause: The scope of this parameter was not correctly understood. Only the first-level paragraph depth was set, and nested physical and chemical data table levels were not covered.
How to Confirm Proper Configuration
- Upload a refractory material document containing a chemical composition table and application scenario images, and check if the parsing result includes image text and table data extracted via OCR.
- Initiate a search containing product grade and physical and chemical parameter keywords, and verify that the field units in the recalled results match the original document.
- Test exporting the full knowledge base and importing it to a new environment, and check if all custom metadata fields are fully retained.
- Adjust the
PARSE_MAX_PARAGRAPH_DEPTHparameter, and verify that nested hierarchical content in long documents is fully recalled.
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.