What the data for this category looks like
Data for this category comes primarily from enterprise databases, data dictionaries, masking tool output results, and access approval systems. It is updated per project batch or single task. Daily produced documents include masking plans, field mapping tables, masking verification reports, and approval records. Most document structures use structured fields, including sensitive field identifiers, masking rule codes, verification results, and approval status. There are no unified fixed units; specific field units are adjusted based on project requirements.
What constraints these characteristics impose on the knowledge base retrieval and recall workflow
Data for this category updates per project batch. Retrieval must strictly limit dataset scope to prevent cross-project sensitive data from being recalled. Documents with a high proportion of structured fields require support for precise field matching. Full-text search alone cannot meet the semantic query needs of structured fields. Masking-related documents involve compliance content. Retrieval results must balance accuracy and safety, while enabling fast location of specific project’s masking configuration records to improve retrieval targeting.
Configuration Settings
This configuration is adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Adapts to structured field matching for masking documents, balances accuracy and recall coverage |
limit | Top 8–12 entries | The number of masking documents for a single project batch is usually limited, avoids returning excessive redundant results |
searchMode | vector | For structured rule-based content, vector retrieval is better suited for semantic matching needs than full-text retrieval |
usingReRank | Enabled | Masking documents have high compliance and accuracy requirements; re-ranking improves the precision of result relevance ranking |
datasetSearchUsingExtensionQuery | Disabled | Masking configuration documents do not require extended queries, avoids introducing masking rules from unrelated projects |
collectionIds | Fill in the dataset ID bound to the project | Limits retrieval scope only to the current project’s masking-related datasets, avoids interference from cross-project data |
Three Common Misconfigurations
- Phenomenon: Retrieval results include non-current project masking documents and approval records. Cause: The
collectionIdsparameter is not correctly bound, and the retrieval dataset scope is not limited, leading to cross-project sensitive data being recalled and violating compliance requirements. - Phenomenon: Masking rule matching accuracy in retrieval results is overly inconsistent, either missing key configurations or including irrelevant content. Cause: The
similarityvalue does not align with the semantic characteristics of structured masking documents, and the value range does not fit the actual project needs. - Phenomenon: Unable to correctly match variable configuration parameters in masking schemes during retrieval. Cause: The retrieval mode or matching rules are not correctly configured, leading to failure to accurately locate document content related to variable assignment, corresponding to the variable reference configuration requirement in technical documentation and API Q&A scenarios.
How to Verify Correct Configuration
- Check the FastGPT retrieval configuration panel, confirm that
collectionIdsis bound to the current project’s masking dataset ID, and no additional irrelevant datasets are selected. - Run a test retrieval, enter keywords from the masking plan, and check if returned results only include current project-related documents, with no cross-project data included.
- Adjust the
similarityvalue range, compare retrieval result relevance across different values, and select a configuration that meets the project’s compliance verification requirements. - After enabling
usingReRank, verify that the ranking of retrieval results follows relevance priority, ensuring key masking rules appear at the top.
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.