What the Data for This Category Looks Like
Data is primarily sourced from internal standardized operation manuals, safety management regulations, ticketing service specifications, emergency response plans, and document template libraries of urban public transit enterprises, as well as public transit operation policy documents released by local transportation authorities. Most documents follow a chapter-based structured format, including fields such as applicable scenarios, execution processes, and responsible entities. They involve specialized units including shifts, passenger flow, and maintenance cycles. Update frequency aligns with policy adjustments or internal revisions. Core documents are reviewed on a quarterly basis.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall
The document characteristics of this category impose multiple constraints on retrieval and recall. Structured chapter-based documents require retrieval to match chapter content for specific target scenarios, to avoid retrieving irrelevant operation records. The multi-field design requires support for precise matching via fields such as responsible entities and applicable scenarios, to improve the targeting of document drafting. The presence of specialized units such as shifts, passenger flow, and maintenance cycles requires retrieval to support semantic matching of terms, to avoid errors caused by unit confusion. The non-fixed update frequency requires recall logic to prioritize filtering policy and template documents updated in the most recent quarter, to avoid using expired content. Additionally, the unified document template format requires retrieved content to retain its original structure, to ensure direct reusability.
How to Set Configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Urban public transit documents are mostly professional compliance content, requiring filtering of low-correlation results while retaining sufficient candidates for document drafting |
limit | Top 8 results | Document drafting requires reference to multiple templates and policies. 8 results can cover different dimensions of common scenarios while controlling context length |
searchMode | Hybrid Retrieval | Balances semantic matching and keyword matching, adapting to the dual retrieval needs of structured document chapter titles and content keywords |
usingReRank | Enabled | Urban public transit documents contain a large number of professional terms. Re-ranking can further filter results highly relevant to the document drafting scenario |
datasetSearchUsingExtensionQuery | Enabled | Can automatically expand query keywords for document drafting. For example, expanding "spring festival travel scheduling" to "spring festival emergency scheduling public transit plan", improving retrieval precision |
collectionIds | Only collections of public transit operation manuals, policy documents, and document templates | Only retrieve datasets directly relevant to urban public transit document drafting, avoiding mixing in irrelevant data |
Three Common Mistakes
- Phenomenon: Retrieval results return a large amount of non-document content such as vehicle maintenance and driver attendance, with unsatisfactory relevance. Cause: The exclusive dataset collection is not restricted, and operation data irrelevant to document drafting is retrieved.
- Phenomenon: After entering "bus spring festival travel emergency plan", only a small number of literal matching results are returned, without covering relevant scenarios such as passenger flow control and station scheduling. Cause: The question optimization function is not enabled, and semantic related keywords of the query are not automatically expanded, so hidden associated content cannot be matched.
- Phenomenon: The latest public transit fare policy document uploaded cannot be retrieved. Cause: The file is not classified into the specified policy document collection, or the original chapter structure of the document is not retained during upload, causing key clauses to fail to be correctly retrieved.
How to Confirm Configurations Are Set Correctly
- Upload one urban public transit document template and one vehicle maintenance record, perform a retrieval test, and confirm that only document template-related results are returned, to verify that the collection restriction configuration is effective.
- Enter a query containing professional terms, such as "application document template for new line opening", confirm that retrieved results include corresponding chapter content, to verify that the question optimization and re-ranking configurations are effective.
- View the dataset collection list in the knowledge base, confirm that only three types of datasets are included: operation manuals, policy documents, and document templates, to verify that the specified collection ID configuration is effective.
- After adjusting retrieval parameters, execute the same retrieval request multiple times, confirm that the number and relevance of returned results remain stable, to verify configuration consistency.
The values given 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.