What this category of data looks like
This category of data primarily comes from IAM systems, HR systems, business system logs, and approval systems. Daily maintained documents include account application forms, permission matrices, terminated account recovery forms, and compliance audit reports. Document fields include account ID, permission scope, approval status, change time, and more. Data updates occur daily or in real time alongside personnel changes. Individual document sizes range from several pages to dozens of pages. Core fields mostly combine structured and semi-structured data. Some permission matrices contain multi-dimensional associated fields.
What constraints do these characteristics impose on knowledge base retrieval and recall
Data updates in real time with personnel changes. This requires knowledge base recalled content to stay synchronized, to avoid using outdated permission rules or terminated account information. Structured and semi-structured document fields demand precise matching of fields such as account ID and permission scope during retrieval, to avoid retrieving irrelevant content. Multi-dimensional associated permission matrix data requires context association during recall, to ensure complete returned permission descriptions. Single retrieval must cover documents from multiple system sources. Retrieval scope must be limited to avoid mixing irrelevant data from other categories.
How to set configurations
This document is written based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Permission-related documents in this category require precise matching to avoid low-similarity irrelevant permission rules |
limit | Top 8–12 entries | Documents such as permission matrices have high content relevance, and a small number of entries can cover complete information |
searchMode | vector + fulltext | This category has both structured fields and semi-structured documents. Hybrid retrieval balances accuracy and recall rate |
usingReRank | Enabled | Multi-dimensional permission data often has entries with similar semantics but different scenarios. Re-ranking improves the accuracy of relevance sorting |
datasetSearchUsingExtensionQuery | Enabled | Personnel change-related documents need to associate context from account application and approval processes. Extended queries can supplement associated information |
collectionIds | Specify the knowledge base collections corresponding to IAM, HR, and approval systems | Limit retrieval scope to avoid mixing irrelevant data from other categories |
Three common mistakes
- Phenomenon: Irrelevant permission information for terminated accounts appears in retrieval results. Cause: Real-time synchronization update rules are not configured, and the knowledge base has not synchronized terminated account recovery form data from the HR system.
- Phenomenon: Account ID variables cannot be referenced when selecting a knowledge base in the search card. Cause: Variable binding rules are not configured, and retrieval parameters are not associated with the account ID field.
- Phenomenon: Tabular datasets cannot be selected when creating a knowledge base. Cause: Tabular dataset upload permission is not enabled, or the uploaded table does not include required fields such as account ID and permission scope.
How to confirm the configuration is correct
- Manually upload a latest terminated account recovery form, trigger knowledge base synchronization, search for content related to terminated accounts, and confirm that results include the latest status.
- Enter a query containing a specific account ID, check whether retrieval results accurately match corresponding permission matrix entries.
- Attempt to upload a tabular dataset, confirm that the tabular type is included in the optional list, and verify that the dataset selection function works properly.
- After enabling re-ranking and extended queries, compare the sorting of retrieval results before and after enabling to confirm that the re-ranking function takes effect.
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.