Data profile for this category
Data for this category originates from government information disclosure platforms, official archive systems, document management systems, departmental business ledgers, and government affairs disclosure-related specifications. Daily documents include government information disclosure application forms, supplementary notice letters, third-party opinion solicitation letters, response letters, service receipts, and administrative reconsideration response materials. Data is updated irregularly, with no fixed synchronization cycle. Most documents are a mix of structured forms and unstructured formal texts. Core fields include applicant information, application matters, response content, and basis for public attribute determination.
Constraints on knowledge base retrieval and recall
Multi-source data sources require retrieval to support cross-dataset calls, covering document content stored across different systems. Irregular data updates require retrieval configurations to adapt to incremental synchronization logic, preventing the recall of expired or unarchived documents. Mixed-structured documents require retrieval models to balance features of structured fields and unstructured texts, ensuring accurate matching of application matter keywords. The core business goal is to confirm whether specified government information exists. This requires recall results to have high confidence, while limiting interference from irrelevant results to avoid impacting review efficiency and accuracy.
Configuration settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches keyword density and semantic relevance of government documents, avoiding overly low values that bring in irrelevant administrative documents, and overly high values that miss valid information matching application matters |
limit | Top 8 entries | Adapts to context length limits for single-file review, avoiding exceeding large model input limits while ensuring coverage of sufficient relevant document fragments |
searchMode | Semantic + keyword hybrid retrieval | Balances structured fields and unstructured response content of government documents, improving matching accuracy for application matters |
usingReRank | Enabled | Performs secondary ranking on initial recall results to filter low-relevance content, meeting the high confidence requirement of the review link |
datasetSearchUsingExtensionQuery | Disabled | Prevents automatic query expansion from deviating from the specific application matters of the applicant, ensuring accurate matching of core applicant demands |
collectionIds | Bind specified datasets by business scenario | Limits retrieval scope to the current department’s government affairs disclosure-related datasets, avoiding interference from cross-department irrelevant data in the review process |
Three common configuration errors
- Issue: An error
fail to create post presigned urlappears when uploading government documents, and the file upload status shows failed. Cause: Cross-origin resource sharing rules were not correctly configured after migrating S3 storage, or write permissions for the storage bucket were not enabled. - Issue: The number of retrieval return results is far lower than expected, with only a small number of relevant documents returned. Cause: The
similaritythreshold was set too high, filtering valid results with semantic relevance, or thelimitparameter value was too low. - Issue: The retrieval scope cannot be limited to the specified dataset, and the default retrieval covers the full knowledge base. Cause: The
collectionIdsparameter was not correctly configured, or the specified dataset ID was not bound in the retrieval request.
How to verify correct configuration
- Upload a single government document, confirm that no errors such as
fail to create post presigned urlappear, and the file status shows completed. - Enter a typical government information disclosure application matter, verify that the recall results include corresponding document fragments, and that the
similarityscore falls within the preset range. - Check the retrieval configuration panel, confirm that parameters such as
usingReRankanddatasetSearchUsingExtensionQuerymatch the preset settings. - Limit the retrieval scope to the specified dataset, enter keywords that do not belong to this dataset, and confirm that no irrelevant results are returned.
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.