What the data for this category looks like
Data for this category comes from the national unified construction project EIA declaration and approval system, ecological environment zoning control results, pollution discharge permit platform, and monitoring data. Daily processed documents include environmental impact assessment reports or report forms, approval applications, public participation statements, technical assessment opinions, EIA approval documents, and similar materials. Data updates have no fixed cycle, and are generated irregularly alongside project declaration and approval progress. Documents contain both structured fields and semi-structured text, including core information such as project category, EIA level, pollution prevention measures, and other key details. Document field formats vary across different sources, with no unified fixed unit standards.
What constraints do these characteristics impose on the knowledge base retrieval and recall workflow
Data from multiple sources causes inconsistent field formats. The retrieval workflow must adapt to document structures exported from different systems to avoid recall failure caused by field matching deviations. Irregularly updated data requires a knowledge base synchronization mechanism that can be triggered flexibly, without relying on fixed-cycle full updates, to adapt to sudden data generation during project advancement. Documents vary widely in length: some reports span thousands of characters, while approval applications are only hundreds of characters long. The retrieval model must support text inputs of different lengths. Some documents contain sensitive information such as public participation content and pollution prevention measures. The retrieval process must enforce access permissions, so only authorized users can access relevant results. EIA approval requires matching project-specific attributes. Retrieved results must be strongly correlated with the specific parameters of the current project, to avoid distracting generalized irrelevant data from interfering with review.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | EIA approval requires strict matching of core project attributes. A threshold that is too low will introduce historical data from unrelated projects, while a threshold that is too high may miss eligible files of the same category |
limit | Top 10 results | EIA review needs to reference multiple historical approvals and technical opinions. 10 results cover the main reference dimensions, and avoid excessive results increasing review workload |
searchMode | Full-text retrieval | EIA documents need to associate multi-dimensional content such as project category and pollution prevention measures. Full-text retrieval can fully match cross-field associated information |
usingReRank | Enabled | Long-text EIA reports contain redundant descriptions. Re-ranking can filter low-relevance results and improve the recall priority of core business information |
datasetSearchUsingExtensionQuery | Enabled | EIA declaration follows fixed industry terminology habits. Query optimization can add standardized expressions, improving recall accuracy across different expression methods |
Three common misconfigurations
- After importing an XLSX-format EIA declaration spreadsheet, it cannot be parsed correctly, and empty results are returned during retrieval. The cause is that FastGPT's table extended retrieval function is not enabled, and the chunking rules for table data are not configured correctly.
- An error occurs when
ada 002is not selected as the embedding model. The cause is that some deployment versions rely on the interface parameters of this model by default, and the configuration is not replaced with a compatible local embedding model. - Content from non-specified knowledge bases is returned in conversations, and irrelevant results cannot be filtered. The cause is that the knowledge base reference range switch is not configured correctly, and the call is not limited to only EIA-class data sources.
How to confirm the configuration is properly set
- Upload a historical EIA approval document, verify that the retrieval results include historical files of the same project category, and check that the number of results matches the configured value.
- Trigger a retrieval for a specific project, check the similarity score of the returned results, and confirm that the score falls within the configured range.
- Check the data source range of the knowledge base, confirm that only EIA-related documents are included, with no cross-category redundant data.
- Import an XLSX-format EIA declaration spreadsheet, verify that the project information in the table can be read correctly during retrieval, with no parsing errors.
The values listed above are common starting points, and should be evaluated against the user’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.