What data looks like for this category
Data related to document and material drafting for shipbuilding mainly comes from classification society specification documents, technical instructions accompanying ship design drawings, block construction process specifications, node acceptance reports, and full ship delivery documents. Data update rhythm adjusts dynamically with construction project nodes. Block process documents are updated during the block construction phase, general assembly process content is updated after the entire ship is docked, and all acceptance materials are updated before delivery. Most documents use a multi-chapter long text structure, including fixed fields such as ship type, classification society number, construction batch, and material grade. Units use industry-specific metrics including millimeters, megapascals, and knots (for speed).
What constraints these characteristics impose on knowledge base retrieval and recall
Shipbuilding document data mostly consists of long-chapter technical documents with high word count per document. This can lead to overly long single retrieved content, which reduces context accuracy. Documents contain precise identification fields such as classification society numbers and material grades, as well as specialized unit terminology. This requires balancing weight adjustments for semantic retrieval and exact matching. Data is managed by project and updated dynamically, with document versions independent across different projects. This requires support for binding knowledge base collections to specific projects to avoid cross-project data confusion. There are many specialized terms, and basic semantic retrieval may not cover variants of exclusive expressions. This requires additional query expansion capabilities.
How to set up configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Shipbuilding documents are mostly professional technical content. A semantic similarity threshold that is too low will introduce irrelevant documents, while a threshold that is too high will miss relevant parameter content. This range is calibrated based on actual testing |
limit | Top 8 results | A single shipbuilding construction document contains multiple types of parameters. Retrieving too many results will exceed the context window, while retrieving too few will fail to cover complete process requirements |
searchMode | hybrid | Balances semantic retrieval and exact matching to adapt to mixed retrieval needs for precise fields and specialized terminology in documents |
usingReRank | Enabled | Ship documents contain many specialized terms. Reranking can filter results that are semantically similar but not a match, improving retrieval accuracy |
datasetSearchUsingExtensionQuery | Enabled | Automatically generates professional expanded queries based on user questions, covering variants of shipbuilding-specific terminology, such as the association between "speed" and "maximum speed" |
collectionIds | Bind the knowledge base collection ID for the corresponding project | Shipbuilding documents are managed by project. Binding the collection ID for the corresponding project avoids cross-project data interference |
Three common errors
- In a local deployment environment, knowledge base retrieval results do not return image index content, and the interface shows empty image fields. The cause is that the path mapping rule for supporting drawing images in shipbuilding documents is not configured. The default index only grabs text content and does not associate image resource links.
- The number of retrieved results does not match the configured
limitparameter, and the actual number of returned results exceeds the set value. The cause is that the result truncation logic after reranking is not turned off. The reranking module automatically supplements relevant results, causing the upper limit to be exceeded. - When a user enters shipbuilding-specific terminology such as "block construction", the retrieved results do not match expectations. The cause is that
datasetSearchUsingExtensionQueryis not enabled, and no expanded query for specialized terms is generated, so synonymous or associated expressions for the term cannot be covered.
How to confirm the configuration is complete
- Upload a single shipbuilding process document, trigger knowledge base indexing, and check whether the image path fields in the document are included in the indexing logs.
- Enter a query statement containing a classification society number, and verify whether the
similarityscore of the retrieval results falls within the set range of 0.72–0.80. - After enabling
usingReRank, compare the number of retrieval results before and after enabling to confirm that the reranking module correctly filters irrelevant results. - Bind the
collectionIdsfor a specified project, enter a cross-project query statement, and confirm that no document content from other projects is returned.
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.