Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
Data for this category comes from the national petition information system, local petition systems, the 12345 citizen hotline platform, grassroots governance platforms, and historical petition archives. The update frequency is daily. Commonly processed documents include petition registration forms, acceptance notices, transfer and assignment letters, handling opinions, review and re-examination applications, resolution plans, and more. These documents are generated across the full process of petition acceptance, handling, review, and re-examination. Fields include requester subject, matter description, handling progress, policy basis, and no extra custom unit fields.
What constraints do these characteristics impose on the "knowledge base retrieval and recall" link
Multiple data sources with daily updates require the retrieval process to support near-real-time incremental synchronization, to avoid retrieving outdated historical archives or unsynchronized latest petition data. Documents cover the full petition process, with significant differences in fields and functions across document types. Differentiated retrieval weights must be set for different document types such as acceptance notices and handling opinions. Legal handling procedures for petition matters require retrieved results to match official process specifications, with no content conflicting with legal procedures. Historical petition archive data accumulates over time. The number of retrieved documents per request must be limited, to avoid context overload affecting subsequent generation steps.
Configuration settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Petition documents have high requirements for semantic matching accuracy. This interval filters irrelevant content while retaining core relevant results |
limit | Top 8 entries | Petition documents need to cover all process links. 8 results balance completeness and context length |
searchMode | local | Most petition data is stored in in-house systems. Local retrieval ensures data privacy and synchronization efficiency |
usingReRank | Enabled | Petition matters involve policies and procedures. Re-ranking improves the matching degree between results and current requests |
datasetSearchUsingExtensionQuery | Enabled | Petition requests are expressed in diverse ways. Extended queries cover synonyms and policy term variants |
collectionIds | Configured according to petition-specific datasets | Restrict retrieval scope to only petition-related documents, avoiding mixing data from other categories |
Three common mistakes
- Phenomenon: Retrieval results reference non-petition-specific documents, or content inconsistent with legal petition procedures. Cause:
collectionIdsis not configured to limit retrieval scope, mixing datasets from other categories. - Phenomenon: Retrieval time for identical requests varies significantly, with interface display time ranging from 1 second to 20 seconds. Cause:
limitis not set to restrict the number of retrieved entries, and the retrieval scope is not accurately limited, with some requests triggering full dataset retrieval. - Phenomenon: Retrieval results are inaccurate, unable to match the policy basis of the current petition matter. Cause:
datasetSearchUsingExtensionQueryis not enabled, failing to cover term variants of petition policies, leading to semantic matching deviations.
How to confirm the configuration is correct
- Check the
collectionIdsconfiguration, confirm that only IDs of petition-specific datasets are included, with no unrelated collections. - Trigger a test retrieval, verify that the document types of returned results are petition-specific documents such as acceptance notices and handling opinions.
- Compare retrieval results with
usingReRankenabled and disabled, confirm that the matching degree of re-ranked results meets expectations. - Check retrieval logs, confirm that single retrieval time stays within a reasonable range, with no sudden long-time requests.
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.