What data looks like for this category
Data for this category comes from government information disclosure platforms, official archive systems, document management systems, departmental business ledgers, and government disclosure-related regulations. Commonly processed documents include government information disclosure application forms, supplementary notice letters, third-party opinion solicitation letters, reply documents, service receipts, administrative reconsideration response materials, and similar items.
Data updates do not follow a fixed cycle, and change based on specific application processing progress, archiving, or policy adjustments. Most documents are structured forms and official papers, with fixed fields including applicant information, application content, public attribute judgment, third-party rights and interests explanation, relief channels, and more.
What constraints do these characteristics impose on knowledge base retrieval and recall
Scattered data sources and non-fixed update cycles require retrieval to cover historical documents synced across multiple systems and real-time updated handling materials, to avoid missing the latest reply or supplementary notice records. Documents contain sensitive, fixed-format fields such as third-party rights and relief channels, so retrieval must accurately match specific information corresponding to the application content, and avoid generalized recall of unrelated content.
Each individual application has a clear retrieval goal: confirming whether specific information exists, and the basis for public attribute judgment. This requires avoiding recalling proactive public information unrelated to the individual case. Retrieval deviations can lead to delayed replies or correction requests, so recall results need high-confidence matching accuracy.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Match the precise content of government information applications, avoid recalling documents unrelated to the application theme, and balance recall rate and accuracy |
limit | Top 6 entries | Cover multiple documents that may be involved in a single application, while controlling the context length to not exceed processing limits |
usingReRank | Enabled | Reorder initially recalled results by relevance to the application content, ensuring the most relevant reply documents and supplementary notice letters are returned first |
datasetSearchUsingExtensionQuery | Disabled | Avoid generating additional query terms to rewrite the precise content of the original application, ensuring that existing question-answer pairs or original document text in the knowledge base are recalled |
collectionIds | Configured per department-specific knowledge base | Limit the retrieval scope to only the documents related to application-for-disclosure from the current department, avoiding interference from irrelevant cross-department data |
Three common mistakes
- Non-target department documents appear in retrieval results. The cause is failure to configure
collectionIdsto limit the retrieval scope, leading to recall of cross-department data. - Output content deviates from the original knowledge base text. The cause is enabling
datasetSearchUsingExtensionQuery: the system generates additional query terms to rewrite the original application content, resulting in matching irrelevant content. - Insufficient recall results. The cause is either the
similaritythreshold being set too high, filtering out qualifying matching results, or thelimitvalue being set too low, failing to recall enough relevant materials.
How to confirm the configuration is correct
- Upload a single known reply document, initiate a query for the corresponding content, and verify that the returned results include the complete original text of this document.
- View the retrieval log to confirm that all sources of recall results are the configured department-specific knowledge base, with no cross-department data included.
- Adjust the
similaritythreshold, initiate a fuzzy matching query, and confirm that the relevance of the recall results meets expectations. - Disable
usingReRank, compare the result order before and after enabling the feature, and confirm that the reordering function operates correctly.
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.