This document is based on FastGPT v4.17.0.
What this category of data looks like
Typical data for this category comes from daily business documents of manufacturing enterprise equipment departments and workshop maintenance teams. These include equipment ledgers, inspection checklists, maintenance plans, work orders, failure analyses, spare parts lists, and special equipment inspection records. Data sources cover EAM/CMMS systems, equipment sensor data, operator inspection records, maintenance history, spare parts inventory, and statutory inspection information. Update frequencies include shifts, daily/weekly/monthly cycles, and real-time updates when faults occur. Document structures include structured tables and unstructured reports. Fields involve specific attributes such as equipment numbers, inspection parameters, fault codes, and spare part models.
What constraints these characteristics place on knowledge base retrieval and recall
Multi-source and heterogeneous data sources require retrieval functions to support cross-system, cross-format content recall, to avoid missing key equipment-related information. Diverse update frequencies require the knowledge base to adapt to different synchronization cycles, ensuring that the latest inspection records and work orders can be retrieved in a timely manner. Mixed document structures (structured tables and unstructured reports) require retrieval logic to adapt to both field matching and semantic understanding, improving recall accuracy for multiple types of documents. The real-time requirements of fault documents require recall results to prioritize recent maintenance records and latest failure analyses, ensuring timeliness of exception handling. Structured documents with multiple fields (such as spare parts lists) require retaining key attributes during chunking, to avoid losing core matching information during retrieval.
How to set the configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the semantic matching accuracy required for documents such as equipment inspection checklists and work orders, filtering out low-relevance recall results |
limit | Top 10 entries | Covers the number of common documents retrieved by workshop maintenance teams in daily operations, reducing extra costs for manual filtering |
searchMode | fullTextRecall | Adapts to the retrieval needs of mixed document types, supporting extraction of key information from equipment ledgers and failure analyses |
usingReRank | Enabled | Resolves the issue of ineffective sorting after long document recall, reordering initial recall results by relevance |
datasetSearchUsingExtensionQuery | Enabled | Optimizes retrieval query phrasing, converting inspection and maintenance requirements into retrieval logic that better matches knowledge base documents |
collectionIds | Configured by workshop/equipment type grouping | Accurately binds knowledge base collections for corresponding equipment categories, avoiding interference from irrelevant documents across workshops or equipment types |
Three common configuration mistakes
- Failing to associate newly created equipment category knowledge bases when calling the chat interface, resulting in empty responses. Cause: The
collectionIdsparameter is not specified in the interface configuration, and the knowledge base collection for the corresponding workshop or equipment type is not bound. - Sorting results of long document recall do not meet business expectations, and the reranking model is not triggered. Cause: The
usingReRankconfiguration is not enabled, and unordered initial recall results are returned directly, leading to chaotic relevance sorting of long content. - When uploading multi-column Excel inspection checklists or spare parts lists, only two columns of content are recognized, reducing retrieval accuracy. Cause: Documents are not chunked by equipment fields, and key multi-column attributes are not retained, preventing complete matching of inspection parameters or spare part information during retrieval.
How to confirm the configuration is correctly set
- Submit a simulated inspection requirement retrieval request, and check whether the returned results include target documents such as inspection checklists and work orders for the corresponding equipment.
- Review the sorting logic of retrieval results, confirm that the relevance of long document results meets business expectations, and verify that the reranking function is active.
- Call the interface to associate the specified workshop knowledge base collection, and check whether the returned results only include documents of the corresponding equipment category, with no cross-category interfering content.
- Upload a multi-column Excel spare parts list, check whether all key fields are retained after knowledge base chunking, and verify that complete spare part models and inventory information can be matched during retrieval.
The values provided are common starting points and should be measured against the reader's own samples.