What this category of data looks like
Data for port and shipping declaration material and file review primarily comes from ship entry and exit declaration forms, container loading and unloading operation documents, customs declaration files, port and shipping supervision archives, and similar sources. Data updates are triggered by operation workflows. Single declaration materials are updated in real time upon submission. Archived files are synced to the knowledge base on a weekly basis. Each individual document has a fixed structure, including fields such as ship name, voyage number, cargo category, declaration number, and inspection record. Some documents include dedicated unit fields like container number, declared weight measured in tons, and assigned berth. Most documents are formatted as structured tables or PDF files with fixed headers.
What constraints these characteristics impose on knowledge base retrieval and recall
Port and shipping declaration materials and files have many structured fields and dedicated identifiers. This requires retrieval to accurately match field-associated content, to avoid mixing in irrelevant results. Real-time updated declaration materials require the knowledge base sync frequency to align with operation windows, to prevent retrieval of expired data. Individual document lengths vary widely, from short declaration entries of tens of characters to hundreds of pages of inspection files. This requires recall matching rules adapted to different text lengths. Fixed field formats require retrieval parameters to be tailored for numeric and alphabetic identifiers, to prevent recall failures caused by format discrepancies.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Aligns with the matching precision required for port and shipping dedicated identifiers. Prevents irrelevant results from being included if set too low, and avoids missing valid files if set too high. |
limit | Top 8 entries | Balances retrieval efficiency and the reference volume needed for a single batch review, and fits standard port and shipping operation review scenarios. |
searchMode | Exact match | Designed for structured fields and dedicated identifiers, to improve the recall accuracy of field-associated content. |
usingReRank | Enabled | Performs secondary ranking on long documents and multi-field content, to optimize recall results for long texts such as inspection files. |
datasetSearchUsingExtensionQuery | Enabled | For short declaration materials, expands matching of associated fields to supplement potentially valid information. |
collectionIds | Configured based on port business scope | Restricts retrieval to the knowledge base collections of specified ports, to avoid recalling irrelevant data across ports. |
Three common mistakes
- Symptom: The number of returned declaration files in retrieval results is insufficient, or a large number of non-local port operation documents are mixed in. Cause:
collectionIdsis not configured to specify the knowledge base collection for the target port, leading to cross-collection recall of irrelevant data. - Symptom: Retrieval results fail to match dedicated identifiers such as container numbers and port operation numbers. Cause:
searchModeis set to full-text match, and the exact match logic for structured fields is not enabled. - Symptom: A format error prompt pops up when uploading port and shipping archived files to the knowledge base. Cause: Documents are not converted to the structured format supported by the platform, or the fixed field headers of port and shipping documents are not retained.
How to confirm the configuration is correct
- Upload one standard port and shipping declaration material to a test dataset, run a retrieval test, and confirm that the returned results include the dedicated field content of this material.
- Adjust the
similaritythreshold, compare recall results across different values, and select the matching precision range that meets business requirements. - After enabling
usingReRank, check the sorting priority of long documents, and confirm that the sorting of inspection file content aligns with business logic. - After configuring
collectionIds, verify that retrieval results only include operation data for the specified port, with no irrelevant cross-port documents.
The values listed here 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.