What the data for this category looks like
Data for this category primarily comes from ERP/WMS systems, barcode or RFID collection devices, purchase work orders, storage location management records, AGV transport systems, and weighing equipment. Daily documents include receiving slips, putaway slips, material picking and return slips, inventory transfer slips, inventory count sheets, lot ledgers, first-in first-out (FIFO) and expiration date control records. Data update frequencies include real-time (such as barcode-based storage intake), daily (such as inventory summaries), and monthly (such as full warehouse inventory reconciliation). Document fields include core information such as material codes, lot numbers, storage location numbers, expiration dates, and unit quantities.
Constraints on Knowledge Base Retrieval and Recall for This Category
The multi-source, decentralized nature of data in this category requires knowledge base retrieval to support mixed retrieval of structured ledgers and unstructured documents. Large differences in data update frequencies require support for two modes: real-time incremental synchronization and scheduled full updates. The need for precise matching of core fields such as lot numbers, expiration dates, and storage locations requires retrieval and recall to combine exact field search and semantic search. The business core of account-physical consistency and lot traceability requires that recall results be linked to the source information of original documents to avoid mismatching risks caused by fuzzy matching.
Configuration Settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Meets the precision requirements for matching core fields such as lot numbers and expiration dates, balances recall coverage and result relevance |
limit | Top 8 entries | Covers multi-material lot information required for complete delivery of a single work order, prevents excessive results from exceeding the context window |
searchMode | mixedRecall | Combines semantic search and exact field search, adapts to mixed data scenarios of structured ledgers and unstructured documents |
usingReRank | Enabled | Improves the sorting accuracy of key fields such as lot numbers and expiration dates, reduces interference from low-relevance results |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands keywords such as material codes and lot rules in process specifications, improves recall coverage |
collectionIds | Bind corresponding warehousing and logistics knowledge base collections by business domain | Isolates document data from different workshops or warehouses, avoids confusion caused by cross-domain retrieval |
Three Common Misconfigurations
- Phenomenon: Retrieval results do not include source traceability information of the original documents, only scattered text fragments are displayed. Cause: Document traceability configuration is not enabled, or source file path and page number parameters are not specified in the retrieval request.
- Phenomenon: After calling the
/api/v1/chat/completionsinterface, uploaded warehousing and logistics documents cannot be retrieved. Cause: ThecollectionIdsof the corresponding knowledge base collection are not bound in the request parameters, or the association configuration between the application and the knowledge base is not completed. - Phenomenon: Retrieval results cannot match exact fields in the lot ledger. Cause: The knowledge base’s file processing model is not used correctly to parse structured table data, leading to incomplete extraction of core fields such as lot numbers and expiration dates.
How to Verify Proper Configuration
- Upload a test document containing a lot ledger and a receiving slip, initiate a test query including "query the storage location of the 3rd batch of raw materials in September 2024", confirm that the returned results include traceability information of the corresponding document.
- Call the
/api/v1/chat/completionsinterface, fill in thecollectionIdsbound to the warehousing knowledge base in the request parameters, verify that the interface can retrieve document content from the corresponding knowledge base. - Adjust the
similaritythreshold andlimitparameter, initiate multiple test queries, confirm that the number and relevance of recall results meet business requirements. - After enabling the
usingReRankconfiguration, compare the result sorting before and after reranking, confirm that the sorting priority of key fields such as lot numbers and expiration dates has been improved.
The values provided are common starting points and should be evaluated against the reader’s own samples.