What data for this category looks like
Typical documents include cotton blending or batching lists, process setup sheets, equipment operation records, yarn inspection reports, grey fabric inspection reports, and defect records. Data sources cover raw fiber inspection, ERP systems, real-time data from spinning and weaving machines, evenness testers, fabric inspection machine detection data, and order specification documents. Updates sync with each batch of raw materials, each shift’s production run, and each piece of fabric produced. Document fields mostly cover fiber indicators, process parameters, defect types and locations, inspection values, and similar details. There is no unified fixed format, and most documents are stored split by production link.
What constraints these characteristics impose on knowledge base retrieval and recall
This category’s data updates frequently and splits data by production node. This requires knowledge base retrieval to support high-frequency incremental synchronization, to avoid recalling outdated process or inspection data. The scattered document structure across multiple production links requires retrieval to associate different document types such as cotton blending, process, and inspection data along the production chain. Diverse field types and lack of a unified format require recall rules to adapt to multiple field categories including fiber indicators, defect descriptions, and equipment parameters. A large number of equipment units leads to easy continuous spread of defects. This requires retrieval to simultaneously associate operation records from upstream and downstream equipment, to narrow traceability scope and improve analysis efficiency.
How to set configurations
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the accuracy of text matching for multiple categories such as defect descriptions and process parameters, and avoids recalling irrelevant inspection records |
limit | Top 6–8 entries | Spinning and weaving have many association items between defects and processes. Limiting recalled entries avoids context overload that disrupts analysis logic |
searchMode | fullTextRecall | Adapts to multi-field, unstructured production documents, and supports cross-document associated retrieval of content from cotton blending, process, inspection, and other links |
usingReRank | Enabled | For time-series data such as equipment operation records and defect records, reranking improves the sorting accuracy of associated results, and prioritizes content most relevant to the query |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval based on keywords from cotton blending sheets and process sheets, covers hidden association items such as fiber indicators and twist, and improves recall coverage |
collectionIds | Configured in groups by production link | Independently associates independent knowledge base collections such as cotton blending and batching, process setup, and inspection records, narrows the retrieval scope and reduces irrelevant results |
Three common misconfigurations
- Phenomenon: Retrieval results include unconfigured knowledge base documents, with no clear error prompt but results that do not conform to production chain logic. Cause: The knowledge base collections to be retrieved are not specified via
collectionIds, resulting in full knowledge base being scanned and recalled. - Phenomenon: The retrieval interface returns a
quote type errorerror. Cause: Passed parameter formats do not meet requirements, such as not passing the floating-point value ofsimilarityper specifications, or not using the correct array format forcollectionIds. - Phenomenon: A query that exactly matches a text block in the knowledge base is entered, but the matching result is not recalled. Cause: The
similaritythreshold is set too high, ordatasetSearchUsingExtensionQueryis not enabled, leading to insufficient keyword expansion that cannot match hidden association items.
How to confirm proper configuration
- Upload a single cotton blending or batching list to the specified knowledge base collection, enter "Cotton blending ratio is unqualified", and check that retrieval results only include relevant content from this document.
- Adjust the
similarityparameter to 0.75, enter "Insufficient yarn strength", and check that the sort order of returned results aligns with the association logic between defects and processes. - After enabling
usingReRank, enter "Continuous defects appear in grey fabric", and check that retrieval results include operation records from upstream and downstream equipment. - Check interface call parameters, confirm that
collectionIdspasses the correct knowledge base collection ID, with no null values or invalid IDs.
The values provided are common starting points and should be measured against the reader's own samples.