What the data for this category looks like
The data related to business handling for rail transit equipment mainly comes from official operation and maintenance manuals, maintenance procedures, and business handling guidance documents. Updates are synchronized with process adjustments or new regulation releases. Most documents are structured content, containing fields such as equipment model, maintenance cycle, handling nodes, required materials, and others. Units include kilometers, hours, unit counts, and similar measurements. The length of individual documents varies widely, ranging from single-page process descriptions to dozens of pages of full-cycle operation and maintenance guides.
Constraints on Knowledge Base Retrieval and Recall
The structured nature of rail transit equipment business data requires precise matching of fields such as equipment model and maintenance cycle during retrieval, to avoid irrelevant results from generic text recall. The wide variation in document length means that long document splitting must maintain chapter integrity, while short documents must avoid excessive chunking that causes semantic breaks. Update cycles change flexibly with process adjustments, so flexible re-indexing cycles must be reserved. The presence of multiple unit fields requires retrieval rules to associate unit terminology, to prevent incorrect results from unit mismatches.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Most rail transit equipment business documents are structured professional content. This range filters low-relevance results while retaining accurately matched maintenance and business process content |
limit | Top 8 entries | A single business document covers multiple handling nodes. Retrieving too many entries will exceed the context window. 8 entries covers the associated information for typical business scenarios |
searchMode | vector | Business data contains a large number of professional terms and structured fields. Vector retrieval can accurately match semantics, which aligns well with the needs of this scenario compared to full-text retrieval logic |
usingReRank | Enabled | The process steps for rail transit equipment have strong logical connections. Re-ranking can adjust the order of retrieved results, prioritizing core handling nodes |
datasetSearchUsingExtensionQuery | Enabled | Business handling processes use fixed phrasing. Extended queries can supplement synonyms, such as matching "apply for" with spare parts documents related to "collect" |
collectionIds | Knowledge base ID for the specified equipment type | Business data for different rail transit equipment models is stored independently. Specifying the ID prevents recall of process content from unrelated models |
Three Common Mistakes
- Symptom: After uploading the same business document, it initially displays 8 chunks, then changes to 13 chunks after a period of time, with duplicate indexed entries. Cause: Duplicate standardized process phrasing exists in the document, and overlap parameter settings during chunking are unreasonable, causing the same text fragments to be split and indexed multiple times.
- Symptom: When retrieving the business process for a specified train model, content from other train models is returned. Cause:
collectionIdsis not configured correctly, resulting in recall of knowledge base data from other equipment models. - Symptom: Illustration descriptions in business documents cannot be retrieved. Cause: Image OCR parsing configuration is not enabled, so process text in illustrations is not included in the knowledge base index.
How to Confirm the Configuration Is Correct
- Upload a rail transit equipment business handling document, review the system’s chunked fragments, and confirm that no duplicate text fragments are split and indexed multiple times.
- Enter business keywords for a specified model, verify that retrieved results only include content related to that equipment model, confirming that the
collectionIdsconfiguration is correct. - Initiate a query containing professional terms, check whether the retrieved results are sorted to prioritize core handling nodes, confirming that the re-ranking configuration is active.
- Upload a business document containing illustrations, verify that retrieval returns process content associated with the illustrations, confirming that the image parsing configuration is enabled.
The values provided are common starting points and should be measured 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.