Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Data sources for internal office and employee services for construction machinery include full equipment maintenance manuals, parts bill of materials (BOM) lists, job safety specifications, equipment ledger CSV files, fault code manuals, and more. Updates occur manually when equipment models are updated, or incremental data is pulled daily from equipment management systems and ERP systems. Documents are split into two categories: long text (such as maintenance steps, safety protocols) and structured tables (such as parts specifications, consumable item lists). Fields include equipment model, serial number, fault code, parameter value, and others. Units include units, hours, millimeters, liters, and other professional engineering units.
What constraints these characteristics impose on knowledge base retrieval and recall
Long text documents lose contextual connections after splitting, leading to fragmented retrieval results for maintenance steps. Structured CSV data has strict field alignment requirements; general text datasets cannot accurately identify structured fields like part numbers and fault codes. Exact matching is critical for items such as fault codes and part numbers; fuzzy matching often returns irrelevant results. Data from different equipment categories is isolated. Retrieval scope must be strictly limited to prevent cross-category data from interfering with final results.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the exact matching needs of fault codes and part numbers, filters low-relevance retrieval results |
limit | Top 10 entries | Matches the retrieval scale of single equipment operation and maintenance data, prevents context window overflow |
searchMode | Exact match priority | Prioritizes exact field value matching for structured field retrieval, reduces mismatches |
usingReRank | Enabled | Performs secondary ranking on segments split from long documents, improves relevance of maintenance step content |
datasetSearchUsingExtensionQuery | Enabled | Automatically adds query extensions related to fields for structured CSV data, improves recall rate of structured content |
collectionIds | Knowledge base IDs bound by equipment category | Limits retrieval scope, avoids cross-category construction machinery data interfering with retrieval results |
Three Common Errors
- Phenomenon: Chinese content displays garbled characters after importing a CSV template, while English content displays normally. Cause: The CSV source file was not saved with UTF-8 encoding, or the correct encoding format was not selected during import.
- Phenomenon: Variable reference fields returned by the knowledge base search module are empty, such as equipment serial numbers failing to load correctly. Cause: The variable parameters for the corresponding knowledge base were not properly bound, or the fields referenced by variables were not configured in the imported structured data.
- Phenomenon: Retrieval results include documents from other construction machinery categories, such as bulldozer results appearing in excavator retrieval requests. Cause:
collectionIdswas not configured to limit the retrieval knowledge base set, leading to the retrieval scope covering all platform construction machinery data.
How to Verify Proper Configuration
- Upload a test structured CSV parts list, use the template import function, and confirm that imported fields match completely with no garbled characters.
- Enter a specific equipment model or fault code, and verify that retrieval results only return documents from the corresponding category, with no cross-category interfering content.
- Check the sorting order of retrieval results, confirm that relevant segments split from long documents appear at the top, to verify the reranking function is active.
- Review the vector database collection list, confirm that construction machinery documents for the current category have been correctly imported and stored as vector embeddings.
The values provided here are common starting points, and should be evaluated against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.