What the Data for This Category Looks Like
This category of data comes from government portals, document management systems, government information disclosure directories, policy repositories, and press conference materials, with daily updates. Daily documents include policy documents, policy interpretation drafts, public disclosure directories, information release approval forms, press releases, and website column update records. Fields include unique document identifiers, public disclosure attributes, confidentiality review status, release channels, affiliated columns, update times, and other relevant fields. Some documents include structured approval process markers and publication time limit requirements.
What Constraints Do These Characteristics Impose on the "Knowledge Base Retrieval and Recall" Link?
Retrieval must support incremental synchronization for daily updated data, and avoid repeated recall of expired documents. Retrieval must adapt to different document structures for mixed storage of multiple document types, including structured public disclosure directories and unstructured policy interpretation drafts. Retrieval must add pre-filtering rules using public disclosure attributes and confidentiality review fields, to prevent classified or internal documents from being recalled for public scenarios. Recall results must have strong relevance to meet ledger summary and report generation requirements, avoiding irrelevant content that disrupts ledger accuracy. Retrieval results must strictly comply with publication time limit requirements due to the proactive public disclosure feature, ensuring all recalled documents fall within the publicly available time range.
How to Configure the Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Match policy documents and ledger entries, filter low-relevance content, and avoid irrelevant documents from being included in summary reports |
limit | Top 10 entries | Balance the richness of recalled information and the readability of the ledger, adapting to the summary requirements of multi-document association |
searchMode | hybrid | Combine keyword and semantic retrieval, adapting to the professional terminology and structured directory characteristics of policy documents |
usingReRank | Enabled | Improve the recall accuracy of long documents, avoiding interference from low-relevance paragraphs in policy interpretation drafts |
datasetSearchUsingExtensionQuery | Enabled | Automatically expand policy-related keywords, covering synonyms and related expressions, and improving recall coverage |
collectionIds | Specify the government public disclosure-specific knowledge base collection | Isolate documents of different categories, ensuring that only policy and public disclosure documents of this category are recalled |
Three Common Misconfigurations
- The generated ledger or report shows knowledge base retrieval reference markers, or displays original retrieval input and output content. This happens when reference display for retrieval results is not disabled, or the workflow lacks a rule to hide references. Original retrieval information is then directly output.
- Recalled documents include classified or undisclosed content. This happens when filtering rules for public attributes and confidentiality levels are not configured, or recall results are not screened by document confidentiality status. Sensitive information is included in the public ledger.
- Single retrieval returns an unexpected number of documents, either too many or too few. This happens when the
limitparameter is not set correctly, or the recall time range is not adjusted for document update frequency. Recall results fail to meet the information density needs of ledger summary.
How to Verify the Configuration Is Correct
- Run a single retrieval test. Confirm all returned documents come from government public disclosure knowledge base collections and meet public attribute requirements.
- Review retrieval result display settings. Confirm reference markers are hidden, or the workflow is set to only output final answers without original retrieval content.
- Simulate daily incremental retrieval. Confirm only same-day updated documents are recalled, with no duplicates or expired content.
- Test the keyword expansion function. Confirm retrieval covers policy synonyms and related expressions, ensuring relevant documents are correctly recalled.
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.