What Data for This Category Consists Of
Core data for this category comes from daily government affairs handling documents from government offices, departmental government information disclosure offices, and other entities. These include government information disclosure application forms, supplementary notice letters, reply letters, service receipts, and similar documents. Data sources cover government information disclosure platforms, agency archive systems, document management systems, departmental business ledgers, and government information disclosure related regulations. Update frequency is irregular, synced with new application handling, archive filing, or regulation updates. Each individual document includes fixed fields such as application items, public attribute judgment basis, and remedy pathway explanations. Its format is tightly bound to government disclosure handling workflows.
Constraints Imposed on the "Knowledge Base Retrieval and Recall" Link
Data sources are scattered across multiple internal systems. Retrieval configurations must support associating multiple dedicated datasets to avoid missing the unit’s core handling basis. Irregular update frequency requires flexible synchronization mechanisms to ensure recalled content aligns with latest government regulations and archive data. Fixed document structure involving statutory wording requires precise matching of government terminology during retrieval to prevent irrelevant content from interfering with judgments. Handling targets are individual case requests from specific applicants. Recalled content must strictly correspond to specific application items, with no generalized matching allowed. Error consequences involve administrative reconsideration or judicial correction. Retrieval configurations must strictly control the relevance and accuracy of recalled results.
How to Configure Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches statutory terms in government documents and precisely aligns with application items to avoid incorrect recall of irrelevant content |
limit | Top 8 entries | Controls the number of recalled results per retrieval, adapts to the limited reference scenario of government documents, and avoids information overload |
searchMode | semantic_search | Meets semantic matching needs for fixed expressions in government affairs, replacing stiff results from pure keyword matching |
usingReRank | Enabled | Improves sorting priority of highly relevant results, ensuring core basis related to public attribute judgment and remedy channels is recalled first |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands relevant government disclosure regulations, historical replies, and other query terms based on application items to cover more comprehensive basis |
collectionIds | Filled according to the actual associated department’s dedicated government affairs dataset | Limits retrieval scope to the unit’s own data, avoiding interference from cross-department irrelevant content |
Three Common Misconfigurations
- Phenomenon: Newly created government affairs datasets always appear in the root directory. Cause: Failed to correctly pass the parent ID parameter when creating the dataset, or the parameter format does not meet system requirements.
- Phenomenon: The knowledge base can query data, but no output is generated after passing the data to the large model. Cause:
datasetSearchUsingExtensionQueryis not enabled. The recalled content does not form an effective association with the current application item, so the large model cannot extract valid information. - Phenomenon: Single knowledge base search takes more than 15 seconds. Cause:
collectionIdsis not configured to limit the retrieval scope, resulting in retrieval covering too many irrelevant datasets, or thesimilaritythreshold is set too low, leading to recall of too many redundant irrelevant results.
How to Verify Proper Configuration
- Log in to the FastGPT backend, enter the configuration page of the target government affairs dataset, and confirm that the
collectionIdsparameter is filled with the ID of the department’s dedicated government affairs dataset, and does not include irrelevant datasets. - Enter a simulated application-based government information disclosure application item to trigger retrieval, and check whether the recalled results include relevant documents such as the department’s public application form template and reply specification, with core basis sorted first.
- View the retrieval task logs, confirm that single retrieval time is within a reasonable range, and verify that the
limitandsimilarityparameter configurations take effect. - Associate the large model to generate a reply document, and check whether the generated content references the recalled knowledge base basis, with no empty output or irrelevant content.
The values provided here 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.