What data looks like for this category
Data sources for shipping port marketing content and operational data include port operation management systems, international shipping schedule query platforms, shipper service manuals, route promotion copy, and official announcements. Update cadences vary significantly:
- International shipping schedules are updated daily
- Cargo movement data is synced hourly
- Marketing materials and announcements are updated irregularly alongside route adjustments and holiday events
Three document structure types exist:
- Structured shipping schedules, with fields including vessel name, voyage number, berthing time, cargo type, and throughput
- Semi-structured service manuals, divided into route introduction and fee standard modules
- Unstructured short video scripts and announcement texts
Units follow international standards: berthing time is marked in UTC+8, throughput is measured in TEU, and cargo types are classified by HS codes.
What constraints these characteristics impose on knowledge base retrieval and recall
The coexistence of structured shipping schedules and semi-structured service manuals requires retrieval systems to support both vector semantic matching and precise keyword matching. This avoids matching deviations for schedule fields that occur when only vector recall is relied on.
Frequently updated shipping schedules and cargo movement data require the knowledge base to support incremental update mechanisms. Full re-parsing causes excessive synchronization delays, which disrupts real-time delivery of marketing content.
Wide variation in document length—from short berthing announcements of tens of characters to thousands-of-word route planning manuals—demands adaptive chunking strategies. This prevents over-splitting of short documents or loss of core semantics in long documents.
A large number of port-specific terms and units require a custom term list. This prevents general-purpose models from misidentifying "TEU" as a common English abbreviation, or confusing "voyage number" with synonymous terms from other industries.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunk length | 800–1200 characters | Balances length differences across shipping port documents, avoids over-splitting short shipping schedule announcements while preserving contextual integrity for long service manuals |
recall count | top 8–12 results | Covers multi-dimensional query scenarios for port marketing content, including route, cargo type, and shipping schedule needs |
similarity threshold | 0.72–0.80 | Filters low-match irrelevant results, adapts to semantic matching accuracy for port-specific terms |
incremental upload toggle | enabled | Adapts to frequent updates of shipping schedules and cargo movement data, reduces synchronization delays from full re-parsing |
parsing timeout | 300 seconds | Meets parsing time requirements for large route planning manuals and bulk shipping schedule tables |
custom term list | add port-specific terms including "TEU", "voyage number", "berthing time" | Prevents general-purpose models from misidentifying core port terminology |
The parameter values provided on this page are common recommended starting points for configuration setup. Actual values are affected by material format, data volume, and business rules. Specific issues require case-by-case analysis, and it is recommended to test against your own samples before finalizing settings.
Three common mistakes
- Symptom: Files cannot be read during conversation upload, while the knowledge base upload process shows no errors. Cause: Local deployment has not enabled local storage permissions for file parsing, or the uploaded CSV shipping schedule table has no structured parsing rules configured.
- Symptom: Conversation retrieval returns no results, but the knowledge base backend test can normally recall content. Cause: Real-time cache synchronization is not enabled during retrieval, or the temporarily adjusted similarity threshold falls outside the range adapted for port terms.
- Symptom: Retrieval returns content unrelated to the query. For example, a query for "October shipping schedules for Shanghai Port" returns non-port-related logistics information. Cause: No port-specific term list has been configured, and the general vector model cannot distinguish industry-specific synonymous terms, leading to incorrect recall.
How to confirm proper configuration
- Upload a real CSV shipping schedule table, check that the parsed chunked content fully retains core fields including vessel name, voyage number, and berthing time.
- Submit a query including "berthing time for XX port XX route", verify that the number of recalled results falls within the set 8–12 range.
- Adjust the similarity threshold to 0.75, submit a query including port-specific terms such as "TEU" and "voyage number", confirm that no low-match irrelevant results are returned.
- Upload an updated shipping schedule file, check that the knowledge base automatically updates the corresponding entries without requiring full re-parsing of all historical files.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.