What the data for this category looks like
The core entities in this category are panel furniture, solid wood furniture, custom home and door enterprises. Daily technical documents include panel cutting diagrams, cutting material lists, edge banding and drilling programs, coating process documents, hardware parts lists, assembly trial and packaging records. Data sources cover order design systems, sheet and wood batch ledgers, CNC production equipment, ERP/MES systems, and sheet moisture content testing equipment. Data updates follow three schedules: real-time synchronization upon order generation, daily synchronization of batch production data, and updates to moisture content information after each batch of material testing. Document fields include panel size, hole position coordinates, sheet specifications, hardware model, process parameters, and more. Most field units are millimeters, cubic meters, pieces, and similar units.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Multi-source and heterogeneous data sources require retrieval to support precise filtering across business sets, to avoid mixing documents from unrelated systems. The high-frequency update rhythm requires the recall mechanism to adapt to incremental synchronization logic, ensuring that the latest sheet batches and order cutting data can be retrieved quickly. The mixed structure of multiple document types—including structured lists and unstructured process descriptions—requires retrieval configuration to balance text and metadata matching across different formats. The business model driven by customized orders requires retrieval to associate order-level metadata, to accurately locate production documents for the corresponding order. The matching requirements for high-precision fields such as panel size and hole positions require that the retrieval similarity threshold be set within a reasonable range, to prevent low-relevance results from interfering with the review process.
How to set the configurations
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches high-precision fields such as panel size and hole position coordinates, filters low-relevance retrieval results |
limit | Top 10 entries | Covers the number of panels for a single customized order, avoids retrieval result overload that affects review efficiency |
searchMode | mixedRecall | Adapts to text-based process documents and metadata text linked to cutting diagrams, balances structured and unstructured content |
usingReRank | Enabled | Reranks retrieval results, prioritizing documents with higher matching degrees for panel size and hardware |
collectionIds | Configured by business module grouping | Distinguishes sets such as cutting data, process documents, ERP ledgers, and others, narrowing the retrieval scope |
datasetSearchUsingExtensionQuery | Enabled | Supplements metadata queries such as order numbers and sheet batch numbers, associating cross-system related documents |
Three common errors
- Search results include documents from non-target business modules, such as coating process documents mixed with cutting data. This occurs when the
collectionIdsparameter is not used to specify the retrieval set scope, leading to retrieval covering unrelated knowledge base sets. - Retrieval results only return text summaries, with no access to complete links for original documents. This occurs when the
datasetSearchUsingExtensionQueryconfiguration is not enabled, or when metadata link information for original documents is not associated. - Image paths in uploaded MD-format technical documents fail to load correctly, with the interface displaying image placeholders or error messages. This occurs when local image files are not synchronized to the knowledge base storage directory, or when reference paths do not adapt to the knowledge base's file storage rules.
How to confirm the configuration is correct
- Perform a test retrieval by entering panel size keywords, then check if returned results only include documents from the target business set.
- Select one retrieval result, click the link to verify that it jumps to the original document page, and confirm that complete content is accessible.
- Upload an MD test document containing local images, then check if images load normally without placeholders or error messages.
- Adjust the
similarityparameter value and run the retrieval again, then observe whether changes in the number of results and their relevance meet expectations.
The values provided are common starting points and should be measured against the reader's own samples.