What the data for this category looks like
Data sources for port and shipping business inquiries include official government service guides, ship operation process specifications, container yard management rules, crew certificate application notices, and temporary operation notices. Data sources are public documents from port authorities and shipping regulatory departments, plus logs from internal operation systems. Update cadence falls into three categories: policy documents are updated quarterly or annually, daily operation process documents are updated monthly, and temporary notices such as route adjustments and port closure alerts are updated in real time. Document structure includes structured fields and unstructured text. Structured fields cover operation type, processing time limit, required materials, and contact channels, with units including workdays, copies, meters, and others. Unstructured text mostly consists of detailed operation instructions in PDF format.
What constraints these characteristics impose on knowledge base retrieval and recall
Mixed structured and unstructured data requires retrieval to support field filtering, to prevent retrieval of irrelevant content. Real-time updated temporary notices require an incremental indexing mechanism to ensure the latest information can be retrieved promptly. Content dense with specialized terminology requires the retrieval model to adapt to shipping domain word vectors, otherwise semantic matching deviations may occur. Multi-classified business data must be divided into knowledge base sets by operation type, otherwise retrieval results will mix unrelated business processes and interfere with user queries.
How to set configurations
These configurations apply to FastGPT v4.17.0, customized for the port and shipping business inquiry scenario.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72-0.80 | Port and shipping data contains a large number of specialized terms. This threshold filters low-relevance general content while retaining accurate operation process information |
limit | Top 10 | Business inquiry queries have clear intent, and 10 retrieved results are sufficient to cover all necessary handling points for users |
searchMode | local | All data sources for port business inquiries are internal or public local documents, so no online extended query content is needed |
usingReRank | Enabled | Documents dense with specialized terms require reranking to optimize sorting results and improve the relevance of information available to users |
datasetSearchUsingExtensionQuery | Disabled | User questions for port business inquiries are usually clearly worded, so no extra query terms are needed to supplement semantics |
collectionIds | Divide sets by port operation type | Configure collection IDs by categories such as inbound and outbound declaration, container operations, crew management, etc., to accurately filter documents from unrelated businesses |
Three common mistakes
- After adjusting the context window size, returned results do not reflect knowledge base content associated with historical conversations. The cause is failing to use historical conversations as a pre-filter condition for retrieval, or incorrect configuration of
datasetSearchUsingExtensionQuerycausing the context to not be included in the retrieval scope. - After uploading a port operation guide PDF, a large number of irrelevant contents such as headers and footers appear in the index. The cause is failing to set text filtering rules to remove non-business-related page auxiliary information.
- Attempting to export the knowledge base file list returns an empty value or an error. The cause is failing to enable the knowledge base file metadata collection function, or failing to configure correct export permissions.
How to confirm configuration effectiveness
- Submit a test query of "What materials are required for ship inbound and outbound declaration". Verify whether returned reference documents originate from the corresponding classified knowledge base set to confirm that the
collectionIdsconfiguration is effective. - Adjust the
similaritythreshold to different ranges, compare changes in result relevance across attempts to confirm that the threshold meets business requirements. - After enabling
usingReRank, compare result sorting before and after reranking to confirm that the reranking function operates correctly, with results related to specialized terms displayed first. - Upload a test port temporary operation notice document, verify that the generated index only contains business content to confirm that text filtering rules are effective.
The values listed here are common starting points, and should be evaluated against the samples of the specific deployment.
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.