What the data for this category looks like
Data for this category comes primarily from the engineering quality and safety supervision system, construction site real-name platform, supervision logs, and test reports. Daily documents include supervision plans, supervision inspection records, hidden danger rectification notices, quality problem handling forms, and completion acceptance supervision reports. Data updates occur weekly. Most documents focus on construction site process management and acceptance steps. Fields include supervision subject, inspection items, rectification closed-loop status, and acceptance conclusion. No additional custom statistical values are included.
What constraints these characteristics impose on knowledge base retrieval and recall
Weekly updated data requires matching synchronization cycles for retrieval and recall data sources. This prevents use of expired supervision inspection records or rectification notices. Mixed structured and semi-structured documents need chunking logic tailored to different document types. This ensures accurate recall of both long texts from supervision logs and short texts from rectification notices. Inquiries about construction sites often reference specific project on-site status. Prioritize recalling documents linked to the same project, and exclude acceptance report data from other projects. The hidden danger rectification closed-loop status field needs precise identification. Ensure retrieval results include records for all relevant statuses, such as pending rectification and rectified.
How to set the configuration
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Basis for This Setting |
|---|---|---|
similarity | 0.75–0.85 | Filter low-relevance on-site inspection records, retain highly matching supervision-related documents |
limit | Top 10 entries | Each document in this category has moderate information volume; too many results will exceed the context carrying limit |
searchMode | vector + keyword | Adapt to mixed structured and semi-structured documents, balance semantic matching and accurate keyword recall |
usingReRank | Enabled | Reorder retrieval results, prioritize displaying documents strongly linked to the inquired project |
datasetSearchUsingExtensionQuery | Enabled | Supplement project-linked peripheral documents through extended queries, cover related content such as supervision plans and acceptance reports |
collectionIds | Determined through actual testing | Bind document collections exclusive to this category, avoid mixing in documents from other housing and urban-rural development sub-categories |
Three Common Misconfigurations
- Phenomenon: Retrieval results only include text content, and image information within documents is not extracted. Reason: Image OCR and vectorization configuration is not enabled, only plain text from documents is processed.
- Phenomenon: Retrieval returns empty results or prompts that no matching documents are found. Reason:
collectionIdsis not configured correctly, and document collections from non-this-category are bound, leading to incorrect retrieval scope. - Phenomenon: Context exceeds the carrying limit, and the large language model cannot complete the response. Reason: The
limitparameter is not set reasonably, too many long text documents are recalled, exceeding the preset context length.
How to Confirm the Configuration Is Correctly Set
- Upload one typical document of this category, such as a hidden danger rectification notice, initiate a test query matching the keywords of this document, and verify whether the retrieval result includes this document.
- View the document list in the knowledge base, confirm that only category-specific documents such as supervision plans and inspection records are included, with no content from other unrelated categories.
- Initiate a query containing a project name, and verify whether the retrieval result is linked to documents such as the completion acceptance supervision report corresponding to this project.
- After enabling
usingReRank, check whether the order of retrieval results prioritizes content directly related to the current inquiry project.
The values provided 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.