What Data Looks Like for This Category
Data for this category primarily comes from data asset inventories, database metadata, business system descriptions, and publicly available regulatory documents. Daily maintained documents include data catalogs, classification and grading rules, important data identification tables, and review minutes. Data catalogs contain asset ID, asset name, affiliated system, and sensitive field lists. Classification and grading rules define grading thresholds and control requirements. Important data identification tables mark data items, identification bases, and classification results. Data update frequency is quarterly or synchronized when the corresponding business system changes.
Constraints on Knowledge Base Retrieval and Recall
The data sources for this category are scattered, and update schedules are inconsistent. It is necessary to adapt to multi-format, multi-structure document content. Retrieval and recall must accurately match field characteristics across different documents. Daily documents contain precise professional terms such as sensitive fields and grading thresholds. Generic matching can easily lead to retrieval results that deviate from business needs. Data updates are synchronized with business system changes. The knowledge base must support incremental updates to avoid retrieving expired classification and grading rules or identification tables. Temporary documents such as review minutes have strict timeliness. Retrieval must prioritize matching the latest document versions to ensure results meet current compliance requirements.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirement for professional terms in this category, avoids irrelevant results caused by too low a threshold, and avoids insufficient recall of compliance documents caused by too high a threshold |
limit | Top 6–8 results | Adapts to the single content length of classification and grading documents, avoids exceeding the context window, and covers multiple review minutes and rule documents |
searchMode | Vector search + keyword retrieval | Balances semantic matching for professional terms and precise matching for keywords, adapting to structured documents such as data catalogs and identification tables |
usingReRank | Enabled | Performs secondary sorting on recall results to filter out content with low relevance to data security classification and grading rules |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands user questions and supplements domain terms such as data security and classification and grading, improving the accuracy of retrieval and recall |
collectionIds | Bind corresponding data security classification and grading knowledge base collections | Limits the retrieval scope, avoids recalling document content from other categories, and ensures retrieval results only cover data from this category |
Three Common Misconfigurations
- Symptom: The conversation interface returns empty knowledge base retrieval results, but entering the knowledge base backend’s retrieval test page and submitting the same question returns matching content normally. Cause:
collectionIdsis not configured to limit the retrieval scope, leading to cross-collection retrieval, or thesimilaritythreshold is set too high, filtering out recall results for some professional documents. - Symptom: The knowledge base returns relevant results even when the question is unrelated to data security classification and grading content. Cause: Keyword retrieval is not used to filter vector search results, or
usingReRankis not enabled for secondary sorting, allowing low-relevance content to be recalled. - Symptom: Mathematical or physical formulas in the knowledge base fail to display or output correctly. Cause: Formula recognition during document parsing is not enabled, or the deployed embedding model does not support vectorization of formula content, preventing formula-related content from being correctly retrieved and recalled.
How to Verify Successful Configuration
- Access the FastGPT v4.17.0 knowledge base management page, and confirm that the bound
collectionIdsonly include knowledge base collections related to data security classification and grading. - Submit a test question related to classification and grading rules, and verify that the
similarityscore of the retrieval results falls within the 0.75–0.85 interval. - After enabling
usingReRankanddatasetSearchUsingExtensionQuery, submit a question containing professional terms, and confirm that recall results prioritize classification and grading rule documents. - Upload a review minute document that includes formulas, submit a relevant test question, and confirm that formula content can be retrieved normally and displayed correctly in responses.
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 2026-10-19.