What data for this category looks like
Data related to construction machinery document and material drafting mainly comes from equipment original manufacturer manuals, industry operation specifications, maintenance and acceptance records, and compliance documents. Data updates follow new model mass production and industry standard revisions, with no fixed cycle. Most document structures are chaptered and structured, including fields such as model number, rated lifting capacity, operating radius, and working condition requirements. Units are mostly physical quantities like tons, meters, and hours. Some documents contain long-text operation process descriptions.
What constraints these characteristics impose on knowledge base retrieval and recall
Construction machinery document data has many structured parameters and strict unit requirements. Retrieval must accurately match fields and units to avoid returning irrelevant documents. A high share of long-text operation process documents increases the computational load of vector encoding and recall. Data updates have no fixed cycle, so systems must support fast switching to documents of the latest models via datasets. Document drafting requires associating exclusive parameters of specific models, so retrieval scope must be limited to the corresponding category’s dataset to prevent mixing in general documents.
How to set the configuration
Reference configurations under FastGPT v4.17.0 are as follows:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Construction machinery parameters have high precision requirements. This interval filters low-match irrelevant documents and retains valid retrieval results |
limit | Top 8 entries | The number of documents related to a single model is limited. Excessive results will exceed the large model context window and affect the coherence of document drafting |
searchMode | vector | Balances matching needs for structured parameters and long-text processes, and adapts to the mixed content structure of construction machinery documents |
usingReRank | Enabled | Long-text documents account for a high proportion. Reranking further improves the relevance of retrieval results and optimizes the content accuracy of document drafting |
datasetSearchUsingExtensionQuery | Disabled | Construction machinery terminology and parameters are fixed. No additional query expansion is needed to avoid introducing irrelevant retrieval results |
collectionIds | Dataset ID corresponding to the construction machinery category | Accurately limits the retrieval scope to ensure only exclusive documents of this category are recalled |
Three common mistakes
- Phenomenon: The knowledge base can query data, but the large model outputs no content. Cause: The
similaritythreshold is set too high, and the number of valid data entries recalled is insufficient to provide enough context support for document drafting. - Phenomenon: A single knowledge base retrieval takes more than 15 seconds. Cause:
usingReRankis not enabled andlimitis set to more than 15 entries, which adds extra computation time for vector sorting and context splicing. - Phenomenon: Retrieval results mix general office documents or mechanical materials from other categories. Cause:
collectionIdsis not configured correctly, and the retrieval dataset scope is not limited, leading to recall of irrelevant documents.
How to confirm the configuration is correct
- Upload a construction machinery manual for a specified model to the corresponding dataset, enter a query containing the model's parameters, and check whether the returned retrieval results only include documents of this category.
- Check the retrieval time to confirm that the results meet the optimized configuration expectations and there is no obvious timeout.
- Trigger the document drafting process, check whether the large model can generate corresponding content based on the recalled documents, and there is no empty output or irrelevant content.
- Compare the retrieval results with and without
usingReRankenabled, and confirm that the top 3 results have higher relevance after enabling.
The values given 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.