What the data for this category looks like
Port and shipping business process data is primarily sourced from official job process manuals, government public service guidelines, vessel operation specifications, and other documents published by individual ports. Data updates are triggered by industry policy adjustments, port operation process optimizations, or new route launches, with no fixed cycle. Most documents are structured text, containing fields such as process item name, handling entity, handling time limit (measured in working days), required material list, consultation contact information. Some documents also include exclusive port attribute content such as berth numbers and route information.
What constraints these characteristics impose on the "knowledge base retrieval and recall" link
Port and shipping business data includes exclusive fields such as job processes differentiated across ports and handling time limits calculated in working days. This requires retrieval to accurately match business scenarios and port attributes to avoid confusion of cross-port processes. Structured content such as material lists and handling steps requires that recalled results fully retain entry structures without fragmented truncation. Data updates have no fixed cycle, so reindexing must be triggered promptly after policy or process adjustments. Otherwise, recalled content will lag behind the latest requirements. Some documents contain narrow-scope exclusive information such as berths and routes. The retrieval scope must be limited to the corresponding port's dataset to prevent irrelevant content from being included.
How to configure the settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Port and shipping business process guidance has high precision requirements for keyword matching. This range filters low-relevance general process content and retains results that accurately match port-specific processes |
limit | Top 8 entries | The number of business process documents for a single port is moderate. 8 entries cover the associated information of major handling scenarios and avoid excessive results interfering with the context |
searchMode | Exact match mode | Adapts to the strong rule attributes of port business handling. Exact match mode avoids recalling irrelevant cross-port processes or general content |
usingReRank | Enabled | Port businesses have many similar process descriptions. Reranking prioritizes the most relevant handling steps for user inquiries and improves recall accuracy |
datasetSearchUsingExtensionQuery | Enabled | Port business handling often involves vague user questions such as "how to apply for berthing". Extended queries supplement synonyms and industry terms to match more complete process documents |
collectionIds | Corresponding target port dataset ID | Port businesses have regional differences. Limiting collection IDs ensures only the handling processes of the corresponding port are recalled, avoiding confusion of cross-port content |
Three common mistakes
- Issue: When importing into the knowledge base, business process documents from different ports are mixed together, and indexes cannot be created classified by preset handling items. Cause: An independent dataset was not created for each port's business documents, and the target collection for import was not specified via
collectionIds. - Issue: When calling the knowledge base chat, after adjusting the context length parameter, the associated content of historical questions does not appear in the returned results, and a large number of irrelevant knowledge base content appears instead. Cause:
collectionIdswas not limited to the exclusive dataset for the current business, and thesimilaritythreshold was set too low, causing irrelevant documents to be recalled first and override the associated information of the historical context. - Issue: The returned result references documents that do not include the port-specific process information for the current inquiry, only returning general process content. Cause: The exact match mode of
searchModewas not enabled, or thesimilaritythreshold was set too high, causing accurately matched exclusive documents to not be recalled.
How to confirm the configuration is correct
- Upload exclusive business process documents for a single port, run a retrieval test by inputting inquiry questions corresponding to that port, and check if the returned results only include content from that port.
- After adjusting the context length parameter, launch a session test that includes historical questions, and confirm that the process content associated with the history is recalled first.
- After enabling the
usingReRankfunction, compare the result ranking before and after enabling, and confirm that the process content most relevant to the current inquiry is ranked first. - View the <Reference> tag in the returned results, and confirm that the referenced documents belong to the currently configured dataset.
The values provided 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.