What the data for this category looks like
Reference only; verify against the target FastGPT release and deployment documentation.
What constraints these characteristics impose on knowledge base retrieval and recall
Scanned PDF documents lack original text. An additional OCR extraction step is required, which increases retrieval preprocessing time. Policy document clause numbers are tightly bound to their applicable scenarios. Adjacent clause context must be retained during recall to ensure compliant responses. Internal operation system updates have no fixed cycle. Manual triggering of knowledge base incremental synchronization must be supported. Differences in applicable equipment category fields require retrieval to filter recall results by categories such as EMUs and urban rail vehicles. This avoids mixing irrelevant policy content into results.
How to set the configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Rail transit policy documents contain many specialized terms. This range balances recall precision and coverage, calibrated based on actual testing |
limit | Top 10 entries | Policy documents have long individual entries. Too many results will exceed the context window. 10 entries ensures core clause recall |
searchMode | semantic | Policy clauses use fixed phrasing. Semantic retrieval matches implicitly associated policy bases, suitable for professional document scenarios |
usingReRank | Enabled | Policy documents have many similar phrasing. Reranking filters semantically duplicate recall results, improving response accuracy |
datasetSearchUsingExtensionQuery | Enabled | Policy questions often include equipment model numbers. Extended queries supplement matching of alternative names or standard names for relevant clauses |
collectionIds | ID set corresponding to the rail transit equipment policy knowledge base | Only recall policy data exclusive to this category, to avoid mixing other industry policies that interfere with results |
Three common errors
- Symptom: Retrieval results return a large number of irrelevant non-rail transit policy contents, and some fields are empty. Cause: The
collectionIdsconfiguration to limit the exclusive knowledge base is not set, causing the recall range to exceed the target data. - Symptom: Retrieval results for scanned PDF documents are empty, or return garbled text. Cause: OCR text extraction configuration is not enabled, so content from policy documents in image format cannot be recognized.
- Symptom: Line break recognition errors occur during knowledge base splitting, confusing the string "\n" with actual line breaks. Cause: Text splitting rules are not configured correctly, and chunks are not split using actual line break characters.
How to confirm the configuration is correct
- Upload a scanned rail transit policy PDF, initiate a query for the corresponding clause, and confirm that accurate text results are returned.
- After configuring
collectionIds, initiate a query for non-rail transit policies, and confirm that no irrelevant results are returned. - Initiate multiple rounds of policy-related question queries, check the retrieval logs, and confirm that
usingReRankanddatasetSearchUsingExtensionQueryare active. - Test batch multi-question queries, confirm that no timeouts or result confusion occur during single retrieval.
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.