What the data for this category looks like
Data sources for internal port and shipping office work include in-port scheduling logs, berth allocation rule documents, crew operation manuals, vessel dynamic reports, in-port equipment maintenance records, and more. Update cycles fall into three categories:
- Real-time scheduling logs update synchronously with ongoing operations
- Vessel dynamic reports update daily
- Static documents such as operation manuals and maintenance specifications update quarterly
Document structures include structured tables (such as berthing schedules, operation duration statistics), long-text operation specifications, PDF-format crew qualification guides, and CSV-format operation data ledgers. Fields include berth number, vessel draft depth, operation duration, crew qualification level, and more. Units use professional measurement standards such as meters, hours, and tons.
What constraints these characteristics impose on knowledge base retrieval and recall
The characteristics of port and shipping data directly constrain the configuration logic of retrieval and recall:
- The update cycles of real-time scheduling logs and static documents differ greatly. Separate knowledge base collections must be used for management to avoid outdated data being included in full retrieval.
- The mixed document structure of structured tables and long text requires balancing keyword matching and semantic matching retrieval modes.
- Professional fields and units require retaining term semantics during retrieval, to avoid loss of field information due to chunk truncation.
- Data volume fluctuates with in-port operation volume. The number of single-recall results must be controlled to avoid exceeding the context capacity limit.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Port operation documents are dense with professional terms. A threshold that is too low will introduce irrelevant content, while a threshold that is too high will miss matching professional expressions. Calibrated based on actual testing. |
limit | Top 8–12 results | The port knowledge base contains multiple types of business data. Too many single recall results will exceed the context length limit, while too few will fail to cover all relevant information. |
searchMode | hybrid | Balances keyword matching and semantic matching to adapt to the mixed retrieval needs of structured tables and long-text specifications. |
usingReRank | Enabled | Port data has high professionality. Basic recall results may have semantic deviations, and reranking can improve the accuracy of relevance ranking. |
datasetSearchUsingExtensionQuery | Enabled | Port operation questions often include professional abbreviations such as TEU and LOA. Extended queries can cover term variants and expand retrieval scope. |
collectionIds | Collection IDs grouped by document type | Distinguishes real-time scheduling data, static operation manuals, and crew qualification documents to avoid cross-category content interfering with retrieval results. |
Three common misconfigurations
- Symptom: After calling the retrieval interface, the
usingReRankfield returns false, and returned results are not sorted via reranking. Cause: The reranking model is not correctly bound to the current knowledge base collection, or the reranking function was not enabled during configuration. - Symptom: The exported dataset uses the standard question-and-answer table format, and original operation documents cannot be uploaded directly. Cause: The question-and-answer dataset template was mistakenly selected during export, and the original document export mode was not chosen.
- Symptom: The number of retrieval results is less than the set
limitvalue, and no error prompt is displayed. Cause: The number of valid chunks in the knowledge base collection that meet thesimilaritythreshold is insufficient, or key paragraphs containing professional terms were truncated during chunking.
How to confirm the configuration is complete
- Open the knowledge base collection configuration panel in FastGPT v4.17.0, and confirm that switches such as
usingReRankanddatasetSearchUsingExtensionQueryare enabled. - Initiate a test retrieval, enter a query containing port professional terms such as "100,000-ton berth scheduling process", and check whether returned results include relevant document chunks.
- Export the current knowledge base configuration log, and confirm that parameters such as
searchModeandlimitmatch preset values. - Trigger a reranking test, compare result sorting differences between enabled and disabled reranking, and confirm the reranking function is active.
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.