What data looks like for this category
The service subjects for this category are petition bureaus, department petition offices, subdistrict/township offices and responsible unit offices. Daily processed documents include standardized forms and documents such as petition registration forms, acceptance notices, transfer and assignment letters, handling opinions, review and reexamination applications, resolution plans and other fixed-format materials. Data mainly comes from national petition information systems, local petition systems, 12345 hotline platforms, grassroots governance platforms and historical petition archives. Latest matters and closed archives are updated daily. Document fields include requester information, matter details, handling process nodes, policy basis and other content, with no additional custom extended fields.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Multi-source data synchronization requires retrieval to support unified recall of cross-system documents. Daily update frequency requires scheduled incremental indexing tasks to ensure data timeliness. Standardized documents and forms must retain field associations to avoid splitting that breaks business logic. Core demands of petition matters must accurately match policy basis and historical handling cases. Low-relevance recall results may lead to repeated petitions or petitions submitted to higher-level authorities than required. Strict filtering of invalid content is required, while recall results must cover all documents related to the complete handling process.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Filters low-relevance results, accurately matches the acceptance scope and policy basis of petition matters, and avoids invalid recall interfering with handling opinion generation |
limit | Top 10 entries | Controls the total number of recall results, adapts to the context length requirements of petition business, and avoids exceeding the model's processing limit |
searchMode | Semantic Retrieval | Adapts to unstructured demand text of petition matters, and accurately matches core content of policy terms and historical handling cases |
usingReRank | Enabled | Performs secondary ranking on initial recall results, prioritizes returning handling rules and historical files directly related to petition demands, and reduces review and correction risks |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval terms based on petition demand text, covering synonyms, policy abbreviations and other content, improving recall coverage |
collectionIds | Specify collections by petition business scenario | Limits the retrieval scope to exclusive documents such as petition registration forms and handling opinions, avoiding mixing in irrelevant business data |
Three common configuration mistakes
- Symptom: Knowledge base status shows not ready after server restart, and retrieval process gets stuck with no response. Cause: Automatic incremental index trigger rules are not configured, only manual index triggering is supported. Restarting the service before completing indexing after importing a large number of files will cause this issue.
- Symptom: Embedded images in imported petition documents cannot be displayed properly. Cause: Image extraction and storage configuration in document parsing is not enabled, resulting in failure to load image content during retrieval.
- Symptom: Knowledge base retrieval takes more than 15 seconds. Cause: The number of recall entries from the
limitparameter is not restricted, or lightweight re-ranking mode is not enabled, causing too many invalid results to participate in sorting and context splicing.
How to confirm the configuration is correct
- Upload a standard petition registration form, initiate a retrieval, and check whether the recall results include the same type of historical files and corresponding policy documents.
- View the knowledge base synchronization logs, confirm that the daily update task is running normally, with no failed error records.
- Simulate submitting a petition demand, test whether the retrieval response time meets the real-time requirements of the business scenario.
- Check whether the expanded terms generated by
datasetSearchUsingExtensionQuerycover common terms used in petition business, such as "review and reexamination", "transfer and assignment" and other terms.
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.