What the Data for This Category Looks Like
Data for this category comes from ERP/MES systems, component batch ledgers, tightening and refrigerant charging equipment logs, safety compliance testers, performance laboratory reports, and after-sales fault records. Daily documents include configured BOMs, assembly work instructions, per-unit refrigerant charging records, electrical safety test reports, performance sampling inspection reports, and repair ledgers. Data update frequencies cover per-unit products, shift-level output, and sampling inspection plan cycles. Document fields include product model, component batch number, test values, safety compliance markings, and category-specific parameters such as cooling capacity, pressure values, and energy efficiency grades.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall
Data for this category is multi-source and updated frequently. Charging records for single units, shift output ledgers, and sampling reports are updated in real time. The knowledge base must support incremental synchronization and high-frequency refreshing to avoid data lag that reduces bid quotation accuracy. Documents contain highly compliant content such as safety compliance tests and performance parameters. Retrieval must accurately match product models and batches to prevent recall of incorrect data related to mixed-line misassembly. Repair records and after-sales fault data are linked to post-bid after-sales cost accounting. The system must support quick recall of historical processing solutions by fault type, while also supporting retrieval adaptation for multi-format documents such as equipment logs and system reports.
How to Configure the System
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Accurately match highly relevant fields such as household appliance product models and safety compliance parameters, filter low-similarity irrelevant documents, and avoid mismatched data in bid documents |
limit | Top 6–10 | Cover multiple types of information required for bidding such as configured BOMs, performance tests, and repair records, control the number of retrieval results to comply with model input limits |
usingReRank | Enabled | Perform secondary sorting on recall results by business priority, prioritize returning safety compliance and core BOM data, and adapt to the content requirements of bid documents |
datasetSearchUsingExtensionQuery | Enabled | Automatically expand retrieval keywords, supplement category-specific terms such as product models and safety compliance grades, and improve recall rates for segmented documents |
collectionIds | Configure dedicated collection IDs grouped by product model | Distinguish knowledge base collections for different appliance models, prevent incorrect recall of data related to mixed-line assembly across models |
Three Common Mistakes
- Phenomenon: Only text is displayed after importing PPT format documents into the knowledge base, with no image content. Cause: Document image extraction configuration is not enabled, or images in the PPT are not embedded in an extractable format.
- Phenomenon: The number of returned results during retrieval does not match the configured
limit, or BOM data of the target model cannot be recalled. Cause:collectionIdsare not configured in groups by product model, leading to mixed retrieval of cross-model data, or thesimilaritythreshold does not meet the category accuracy requirements. - Phenomenon: Image link content configured in the knowledge base cannot be output during conversation. Cause: Image resources have not been converted to a public URL format recognizable by the model, or the URL link has access permission restrictions.
How to Confirm Proper Configuration
- Manually enter the model parameters of a single household appliance as the retrieval term, and check whether the recall results include exclusive documents such as the BOM and safety compliance test report of the corresponding product.
- Adjust the
similaritythreshold, verify whether the matching accuracy of the recall results meets business requirements, and confirm that no irrelevant cross-model data is mixed in. - Import a test PPT document, check whether the image content and text information in it can be correctly extracted and recalled.
- After configuring dedicated
collectionIds, retrieve the product parameters of another model, and verify that only knowledge base content of the corresponding model is recalled.
The values provided are common starting points and should be measured against the reader's own samples.