What the data for this category looks like
Document data for rail transit equipment mainly comes from technical manuals accompanying vehicle design drawings, process specifications, type test reports, daily maintenance manuals, and rail industry compliance documents. Data updates trigger when new vehicle projects start, processes iterate, or compliance standards are revised, with no fixed cycle. Individual documents are mostly long texts, with a fixed structure including fields such as model identifiers, component numbers, technical parameters, applicable operating conditions, and maintenance cycles. Units mostly use engineering standard units such as millimeters, newton-meters, and kilometers per hour.
What constraints these characteristics impose on knowledge base retrieval and recall
The long-text structure of rail transit equipment documents requires knowledge base chunks to retain cross-chunk semantic association. This avoids incomplete content caused by fragmented retrieved segments. Fixed fields and professional engineering units require retrieval to support field-level precise matching. This prevents interference from irrelevant parameters or unit ambiguities in results. The authoritative nature of compliance documents requires retrieved results to prioritize official standard documents, to ensure content compliance. No fixed update cycle requires the knowledge base to support flexible incremental synchronization and full update mechanisms. This adapts to document update demands from new vehicles and process iterations. Dense professional terminology further requires the retrieval link to strengthen semantic matching accuracy, reducing the retrieval of irrelevant content.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Rail transit equipment documents contain many professional terms. This threshold filters low-match irrelevant content and ensures the relevance of retrieved content |
limit | Top 8–12 entries | Long text documents generate a large number of chunks after splitting. Excessive results will exceed the context window and affect the completeness of generated content |
searchMode | Hybrid retrieval | Supports both semantic matching and field-level precise matching, adapting to mixed retrieval demands for professional fields and terminology |
usingReRank | Enabled | Semantic similarity calculations for professional documents are easily disrupted by terminology. Reranking improves the ranking priority of highly relevant results |
datasetSearchUsingExtensionQuery | Enabled | Adapts to the extended query function of FastGPT v4.17.0 and above, covering matching for synonymous terms such as "kilometers per hour" and "km per hour" |
collectionIds | Configured by grouping by vehicle model / component type | Narrows the retrieval scope, prevents irrelevant documents across vehicle models from being retrieved, and improves retrieval efficiency |
Three Common Configuration Mistakes
- Calling semantic retrieval and full-text retrieval tools returns a "Connection error" error. The cause is that network access permissions for the Milvus vector database are not configured correctly, or the vector database instance has insufficient resources leading to connection interruption.
- The number of knowledge bases that can be created per single instance is limited, which cannot meet the storage demands for documents of multiple vehicle models. The cause is that FastGPT database configuration parameters are not adjusted. The default pgvector instance limits the upper limit of knowledge base creation.
- Retrieved results include document content from non-target vehicle models. The cause is that
collectionIdsis not configured to limit the retrieval scope, leading to irrelevant documents across vehicle models being included in the retrieval pool.
How to Verify Successful Configuration
- Input typical professional terms from a single rail transit equipment document, and check whether retrieved results include matching content for the corresponding fields.
- Check the sorting logic of retrieval results, confirm that highly relevant content is displayed first after
usingReRankis enabled. - Upload a newly added compliance document, and verify that the knowledge base can complete synchronization according to the configured update mechanism.
- Adjust the
similaritythreshold, and verify whether the number of retrieved results changes as expected along with the threshold.
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.