Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
Data for this category is sourced from the engineering construction project approval management system, construction drawing review system, enterprise qualification database, and public credit information platform, with a daily update frequency. Commonly processed documents include construction permit application materials, construction drawing review qualification certificates, quality and safety supervision procedure documents, completion record forms, and more. These documents contain both structured forms and compliance explanation content, including fields such as participating construction unit qualifications, project approval nodes, and licensing conditions. Some documents include multi-page attachments, and there is no uniform fixed single-document length.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
The data sources for this category are scattered and updated daily, which requires the knowledge base to support incremental synchronization mechanisms to avoid data lag that affects approval efficiency. Documents contain both structured fields and compliance explanation content, so retrieval needs to cover both precise keyword matching and semantic association recall. Core retrieval points focus on key information such as participating unit qualifications and project approval nodes, which places high requirements on recall accuracy. Omitting key approval materials or qualification information may lead to approval delays or compliance risks. The existence of multi-page attachment documents also requires the retrieval strategy to adapt to long-text content, avoiding semantic segmentation breaks that cause loss of key information.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | This category requires precise matching of core fields such as qualifications and approval nodes. A threshold that is too low will introduce irrelevant results, while a threshold that is too high may miss valid approval materials |
limit | Top 8–12 results | Engineering approval materials are mostly batch-associated documents. Retrieving an appropriate number of results at one time can cover all supporting materials required for approval, while avoiding excessive results that increase manual screening costs |
searchMode | semantic+keyword | Balances keyword matching for structured fields and semantic recall for compliance explanations, adapting to the retrieval needs of this category's mixed-format documents |
usingReRank | Enabled | Core information of this category requires precise sorting. Reranking can improve the display priority of highly relevant results and reduce manual verification workload |
datasetSearchUsingExtensionQuery | Enabled | Approval materials often contain professional terms and fixed formats. Extended queries can supplement synonymous expressions, improving recall rates for cross-term retrieval |
collectionIds | Bind to the exclusive engineering approval knowledge base | Isolates knowledge base data from different business lines, avoiding interference from irrelevant documents on retrieval results and ensuring the exclusivity of approval data |
Three common mistakes
- The issue is that retrieval results do not include uploaded approval attachment images. The cause is that the knowledge base image content extraction configuration is not enabled, so the text information in the images is not included in the retrievable scope.
- The issue is that retrieval matching performance for Excel-format approval table data is poor. The cause is that an appropriate indexing strategy is not configured for structured table content, and directly uploading the original table leads to semantic association breaks.
- The issue is that custom indexing does not take effect, and retrieval results cannot hit preset custom fields. The cause is that the custom indexing is not associated with the exclusive engineering approval knowledge base, or the retrieval switch for the custom indexing is not enabled.
How to confirm the configuration is complete
- Upload a single typical approval document, initiate a retrieval targeting the core fields within the document, and check whether the target document is included in the recall results.
- Adjust the
similaritythreshold to verify whether the number and relevance of recall results meet expectations under different thresholds. - Upload structured Excel tables and image attachments, initiate corresponding retrieval requests, and confirm that structured content and image text can be normally recalled.
- Check the dataset indexing status to confirm that the custom indexing and exclusive knowledge base have been correctly associated.
The values provided here 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.