What the data for this category looks like
Port and shipping ledger data primarily comes from terminal operation management systems, vessel scheduling platforms, cargo handling documents, port service fee vouchers, and customs declaration filings. Update cycles range from real-time operation records to daily batch-archived documents. Document structure includes two categories: structured fields and unstructured notes. Structured fields include vessel name, voyage number, cargo handling volume (unit: tons or standard TEU), operation time, operation berth, cargo type, and similar items. Unstructured content mostly consists of operation exception explanations, special requirements, and similar notes. Document formats include CSV, Excel spreadsheets, transcribed text from PDF scans, and structured API exported data.
What constraints these characteristics impose on the knowledge base retrieval and recall workflow
Multi-source and heterogeneous data sources require the retrieval system to support both structured field matching and unstructured semantic recall, to avoid information confusion across data sources. Frequently updated operation data requires the knowledge base to support incremental synchronization, to prevent recalling expired or unarchived ledger content. Fields come with specific units and business identifiers such as TEU and voyage number, requiring the retrieval process to retain field association relationships, to prevent loss of business context from split text. Diverse document formats require the preprocessing link to adapt to different parsing rules, to avoid format disorder in transcribed PDF text that affects retrieval matching accuracy.
How to Configure Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Port ledgers have high business accuracy requirements. This range can filter irrelevant voyages and operation data while retaining valid matching results |
limit | Top 10 entries | Ledger summary and report generation need to cover relevant operation batches. Excessive results will increase context length and affect generation efficiency |
searchMode | Vector retrieval + keyword retrieval | Balances semantic matching and precise matching of structured keywords such as voyage numbers and vessel names, adapting to the multi-field characteristics of ledgers |
usingReRank | Enabled | Port ledgers have a large number of fields. Reranking can improve the ranking priority of highly relevant results and optimize recall quality |
datasetSearchUsingExtensionQuery | Enabled | Ledger queries often include implicit extended information such as voyages and time ranges. This configuration can automatically optimize query conditions and expand the effective recall scope |
collectionIds | Specify the collection ID corresponding to the port operation ledger | Isolate knowledge base data from other categories to avoid interference from non-target documents on retrieval results |
Three Common Misconfigurations
- Context-referenced text contains mixed Markdown markers such as
|and#, which cannot be displayed normally. The cause is that the imported original ledger documents retain the Markdown formatting from editing, and have not undergone format cleaning steps, resulting in retrieval return contexts carrying unrendered syntax symbols. - Ledger queries for specific voyages cannot recall cargo handling data for the corresponding port. The cause is that the keyword retrieval mode of
searchModeis not enabled, and only semantic recall is relied on, which cannot accurately match structured fields such as voyage numbers and vessel names. - A 60-second timeout error is triggered when switching the knowledge base index. The cause is that the configured
limitparameter value is too large, and no time filtering rule is set for the dataset, causing the system to load all historical ledger data and exceed the processing time limit.
How to Verify Proper Configuration
- Import a test port ledger document, initiate a query containing a specific voyage number, and check that the context references have no unrendered Markdown markers.
- Initiate a query for a specific operation berth, confirm that the recall results include handling data for the corresponding berth, and verify that the keyword matching function of
searchModeis active. - Trigger the knowledge base index switching operation multiple times, confirm that no 60-second timeout error occurs, and verify the rationality of the
limitparameter and dataset filtering rules. - Initiate a fuzzy query such as "last week's container handling volume", check that ledger data matching the time range is recalled, and verify that the extended query function of
datasetSearchUsingExtensionQueryis 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.