What this category of data looks like
Data primarily comes from new energy project declaration systems, environmental impact assessment approval archives, grid connection acceptance files, and land use preliminary examination documents. Update frequency adjusts based on project progress: weekly updates during the project preparation phase, monthly updates during the construction and operation phases. Most documents use fixed template formats, with modules including project overview, investment details, and technical parameters. Fields include installed capacity (unit MW), grid connection time, and approval document number. Individual document lengths range from thousands to tens of thousands of characters. Some documents include scanned attachment files.
What constraints these characteristics bring to the knowledge base retrieval and recall link
The wide span of individual document lengths, from thousands to tens of thousands of characters, requires adapting chunking strategies to avoid context truncation or redundant recall. Fields have clear physical units such as MW and approval document numbers. Retrieval must associate unit matching, otherwise non-corresponding category parameter data will be recalled. Update frequency fluctuates with project progress. Incremental synchronization triggered by project nodes must be supported to avoid knowledge base data lag. Some documents include scanned attachment files. The upload process must retain image association information to avoid losing associated content during retrieval.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Meets the accuracy requirement for matching new energy project parameters, avoids recalling irrelevant data |
limit | Top 10 entries | Controls the number of files and declaration materials recalled per single retrieval, adapts to context length limits |
searchMode | vector+keyword | Combines semantic and keyword matching to cover retrieval of fields with clear document numbers and units |
usingReRank | Enabled | Rearranges recalled results by relevance to optimize recall ranking after long document chunking |
datasetSearchUsingExtensionQuery | Enabled | Generates extended queries based on the original question to match implicit technical parameters in declaration materials |
collectionIds | Bound by project category | Limits the retrieval scope to avoid interference from irrelevant data across projects |
Three Common Mistakes
- After uploading docx-format declaration materials and files, attached images cannot be displayed in knowledge base retrieval results. This occurs because the upload process does not retain the associated metadata of images embedded in the document, and only extracts text content.
- Extended recommended questions returned during retrieval do not match the declaration material content in the knowledge base, and are mostly generic AI-generated questions. This occurs because directional extension of
datasetSearchUsingExtensionQueryis not enabled, or the extension scope is not limited to the current project's knowledge base collection. - An error prompt indicating that the model does not exist or cannot be accessed pops up during retrieval, with the identifier aiproxy: 1742438308259968. This occurs because the bound retrieval model has not been deployed or authorized in the current platform instance, and the configured model identifier does not match the actually deployed model.
How to Confirm the Configuration Is Correct
- Upload a test new energy project declaration document, initiate a retrieval request that includes project parameters, and verify whether the returned results include the corresponding field content from the document.
- Trigger the retrieval extended query function, and check whether the recommended questions generated by the system focus on the current project's declaration materials and file review topics, with no generic irrelevant content.
- Compare the sorting of retrieval results when
usingReRankis enabled and disabled, and confirm that the relevance sorting logic of recalled results meets expectations. - Verify the bound knowledge base collection scope, and confirm that it only includes the current new energy project's declaration materials and file data, with no irrelevant content from other projects included.
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.