What Data in This Category Looks Like
This category of data primarily comes from the regulatory and rule repository, higher-level law database, departmental policy document repository, and administrative regulatory document filing system, with irregular update frequencies. Daily processed documents include draft regulatory documents, drafting explanations, legality review opinions, summary of solicited comments, filing reports, and more. Most document structures include provisions, issuing authority, effective date, and review opinion fields. Some documents include public feedback entries from solicited comments, and there is no unified fixed single-document length standard.
What Constraints These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
Dispersed data sources and non-fixed update cycles require the retrieval component to support on-demand synchronization across multiple knowledge bases, to avoid retrieving outdated or invalid regulatory documents. Diverse document types and specialized legal provisions require retrieval logic that balances semantic and keyword matching, to adapt to the specialized text characteristics of regulatory documents. A single review requires comparing multiple types of data such as higher-level laws and departmental policies. The recall upper limit must accommodate multi-dimensional matching needs, while filtering outdated content unrelated to the current review document. Public feedback documents have large field differences, so keyword matching for custom fields must be supported to ensure retrieved opinion content is directly relevant to review points.
How to Configure
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Adapts to the rigorous semantic matching of legal provisions, filtering low-relevance non-regulatory document content |
limit | Top 10 entries | Meets the requirement of comparing multiple types of higher-level laws and policies for a single review, avoiding retrieving excessive redundant data |
searchMode | Semantic + keyword hybrid retrieval | Balances semantic matching for legal provisions and precise keyword positioning, adapting to the specialized text characteristics of regulatory documents |
usingReRank | Enabled | Performs secondary ranking on initial recall results, improving the ranking of highly relevant legal provisions and policy documents |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval terms based on review points, covering synonymous legal terms and policy references with different phrasing |
collectionIds | Bind corresponding knowledge base collections by review scenario | Precise retrieval scope limitation, avoiding introducing data from other fields unrelated to the current review |
Three Common Configuration Errors
- A 403 Forbidden error occurs during debugging. The cause is that the role permissions of the current reviewer are not bound to the associated knowledge base collection, or the access permissions of the knowledge base collection are not opened to the corresponding business department.
- The number of retrieval results is far lower than expected, with only 1 to 2 entries returned. The cause is that the set
similaritythreshold is too high, filtering out some eligible higher-level laws or policy documents. - Retrieved documents include revoked regulatory documents. The cause is that incremental update synchronization for the knowledge base is not configured, and invalid files are not removed from the filing system in a timely manner.
How to Verify Proper Configuration
- Navigate to the knowledge base management page, confirm that the collection corresponding to the bound
collectionIdsfor the current scenario includes the specified data sources such as the regulatory and rule repository and higher-level law database. - Submit a test review point, check whether the retrieval results include regulatory documents and higher-level law provisions from the corresponding field, and that no Forbidden access errors are returned.
- Adjust the
similaritythreshold to 0.8, compare the quantity and relevance of recall results before and after, confirm that the threshold matches the business rigor requirements. - After enabling
usingReRank, check whether the top 3 initial recall results are the legal provisions or policy documents most relevant to the review point.
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.