What This Category's Data Looks Like
The main entities of this segment's data are petition bureaus, department petition offices, subdistrict and township offices, and responsible unit offices. Daily documents include petition registration forms, acceptance notices, transfer and assignment letters, handling opinions, review and re-examination applications, resolution plans, and similar materials. Data sources include the national petition information system, local petition systems, the 12345 citizen hotline platform, grassroots governance platforms, and historical petition archives. Update frequency is daily. Data fields focus on petition demands, handling processes, and review results. There is no unified fixed format. Some documents are structured forms, while others are text-based explanatory materials.
Constraints on Knowledge Base Retrieval and Recall From These Characteristics
Petition data is sourced from scattered locations and updated daily. The retrieval link must support fast access to multi-system synchronized data to avoid data lag. Document types are diverse, including structured form fields and long-form policy explanations. The system must adapt to content matching across different structures. Petition demands are often expressed in colloquial language from the public, which differs from standardized document fields. The system must balance precise matching and coverage. Retrieval results directly affect petition identification and the formation of handling opinions. The accuracy of recalled content directly relates to handling quality. Deviations in recalled content may trigger repeated petitions or petitions submitted beyond the appropriate hierarchy.
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 | There are large differences in petition demand expressions, so balance recall coverage and content accuracy, calibrated based on actual testing |
limit | Top 8 entries | Petition handling requires balancing policy basis and historical cases; 8 entries cover information needs for most scenarios |
usingReRank | Enabled | Petition documents have obvious structural differences; re-ranking improves the sorting priority of highly relevant content |
datasetSearchUsingExtensionQuery | Enabled | Petition demands are often colloquial; extended queries can match standardized document field content |
searchMode | Full-text search | Adapts to content matching needs for multiple document types such as petition registration forms and handling opinions |
Three Common Misconfigurations
- Uploading large PDF documents to the knowledge base results in low accuracy for Q&A related to petitions. The reason is that most documents in petition scenarios are structured forms or short explanatory materials; large chunk size loses core field association information, making precise matching of demand content impossible.
- Setting the knowledge base chunk size to 5000 tokens and the context reference limit to 1500 tokens still results in content truncation errors after retrieval. The reason is that no pre-truncation processing is done on chunked content; the complete retrieved chunk exceeds the reference limit, causing the system to automatically truncate key information.
- Not enabling
datasetSearchUsingExtensionQuery, resulting in colloquial petition demands failing to match standardized policy documents. The reason is that no extended optimization is performed on user demands, making it impossible to cover standardized field expressions in documents.
How to Verify Correct Configuration
- Upload a single petition registration form document, initiate a query matching public petition demands, and verify that the recalled results include the core field content of the document.
- Enter the knowledge base retrieval configuration panel, confirm that
usingReRankis enabled anddatasetSearchUsingExtensionQueryis activated. - Adjust the
similaritythreshold, verify that the number of recalled entries changes as expected with the threshold; a lower threshold results in more recalled entries. - Initiate a query using colloquial petition demand expressions, confirm that the recalled results include standardized policy document content.
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.