What data for this category looks like
Typical data for this category comes from ERP/WMS systems, barcode/RFID equipment, purchase work orders, storage location management, AGV transport systems and weighing equipment. Daily documents include receiving slips, putaway slips, material issue and return slips, stock transfer slips, inventory count sheets, lot ledgers, first-in first-out (FIFO) and expiration date records. Fields cover material codes, lot numbers, storage location numbers, inventory quantities, inspection statuses, expiration dates and more. Units include pieces, kilograms, cubic meters, days and others. Data update frequencies fall into three categories: real-time (such as AGV logistics tracks and weighing data), daily (such as material issue and return slip summaries) and monthly inventory (such as inventory reconciliation sheets).
What constraints do these characteristics impose on knowledge base retrieval and recall
Data for this category comes from multiple sources with widely varying update frequencies. This requires retrieval systems to support multi-source data import and synchronization strategies that combine incremental and full updates. Most documents are structured forms and ledgers, with a large number of precision-matching fields such as lot numbers and storage locations. This requires retrieval to balance semantic understanding and precise field matching. The most time-consuming verification steps involve material identity, lot numbers and inspection statuses. This requires retrieval to quickly recall associated ledgers, and avoid redundant results interfering with core information. Cross-process material location management needs require retrieval to support context queries associated with work orders and processes, to ensure consistent ledger and physical records and complete traceability of lot flow direction.
Configuration Settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.80 | This category’s data mostly consists of structured forms. This range balances recall precision and coverage, avoiding missed matches for key fields such as lot numbers and storage locations |
limit | Top 8–12 entries | Warehousing and logistics ledger data has many entries. Too many results increase model processing load, while too few may miss records for complete sets of required materials |
searchMode | mixedRecall | This category has both structured fields such as material codes and lot numbers, and unstructured content such as expiration date notes and inventory remarks. Hybrid retrieval balances precise matching and semantic understanding |
usingReRank | Enabled | Multi-source data recall may produce redundant results. Reranking prioritizes batch and storage location information related to production ledger summary |
datasetSearchUsingExtensionQuery | Enabled | Queries for production ledgers often require associated extended data such as work orders and AGV logistics tracks. Extended queries supplement associated context |
collectionIds | Configure dedicated collection IDs for warehousing, lot tracking and in-plant logistics separately | This category’s data is divided by business modules. Dedicated collections avoid irrelevant data interfering with retrieval results |
Three Common Configuration Mistakes
- Symptom: Semantic retrieval returns a batch ledger with a score of 0.77, but the final model response indicates no relevant information is found. Cause:
datasetSearchUsingExtensionQueryis not enabled. The system cannot associate extended data such as work orders and storage locations, and cannot match query requirements for ledger summary. - Symptom: A single semantic retrieval returns more than 15 material ledger entries from different storage locations, preventing quick location of the complete set of material information required for the current work order. Cause: The
limitparameter is not configured correctly, orusingReRankis not enabled to prioritize results. - Symptom: After uploading Excel-format material issue and return slips or inventory count sheets, the corresponding file content cannot be found in the knowledge base retrieval interface. Cause: Excel file parsing adaptation is not enabled, or the file is not imported into a dedicated warehousing and logistics data collection.
How to Confirm Proper Configuration
- Check that the
similarityparameter value falls within the 0.72–0.80 range. Run a test query for the ledger with lot number “L20240512001” to verify that recall result matching meets expectations. - Verify the
limitparameter configuration. Run batch retrieval for material ledgers associated with work orders, and confirm that the number of returned results falls within the 8–12 range. - Test the
usingReRankfunction. Upload multiple identical types of inventory count sheets, and confirm that retrieval results prioritize batch expiration date records most relevant to the current query. - Confirm that
collectionIdsis configured with dedicated warehousing and logistics data collections. After uploading an Excel-format material issue and return slip, use keyword retrieval to quickly recall the corresponding record.
Note: The configuration values provided are common starting points. They should be evaluated against respective specific samples and operational scenarios.