What This Category of Data Looks Like
Data for this category originates from server configurations, middleware configurations, security device policies, and baseline tool output results. Daily documents include baseline checklists, configuration verification records, rectification suggestions, and exception approval forms. Fields cover configuration item identifiers, compliance status, verification time, rectification measures, and more. The data update frequency is monthly or quarterly. A single update covers all configuration items and associated approval records for the current compliance verification cycle.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall
This category of data consists primarily of structured configuration items, paired with semi-structured approval and rectification documents. It has a large number of fields and a long update cycle. Retrieval requires precise matching of configuration item identifiers and names to avoid generic recall of irrelevant content. Daily work involves cross-checking configuration items and exception reasons, so recall results must include corresponding rectification suggestions and approval records. Control the context length of single-round recall to avoid exceeding processing limits. Additionally, separate configuration compliance status results from retrieval results of other security services, to keep retrieval results focused on baseline compliance content.
How to Configure Settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | This category has high precision requirements for configuration item fields. This interval filters low-correlation non-compliant configuration retrieval results, matching the precision needs of security baseline verification |
limit | Top 8 | Too many single-round recall results cause context overload. 8 items cover associated configuration items and rectification suggestions for typical security baseline checks, avoiding exceeding processing limits |
searchMode | vector | This category of data is primarily structured configuration items. Vector retrieval can precisely match configuration item names and compliance status, adapting to precise retrieval needs |
usingReRank | Enabled | This category has a large number of similar configuration item names. Re-ranking further filters low-correlation results, improving compliance matching of retrieval results |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval keywords to cover aliases and compliance descriptions of configuration items, adapting to matching of configuration items with different expressions in security baseline checks |
collectionIds | Bind to the dedicated security baseline check dataset | Isolates this category of data from other categories of knowledge base content, preventing irrelevant security service documents from being included in retrieval results |
Three Common Configuration Mistakes
- Semantic retrieval scores display values above 4000. This occurs because the FastGPT score calculation logic is not correctly adapted, and raw vector distance is mistakenly returned as a normalized score.
- Source file information associated with the knowledge base cannot be obtained when calling the API. This occurs because the source file traceability switch for retrieval results is not enabled, and relevant settings for returning source files are not configured in retrieval parameters.
- Some security baseline configuration items return no matching results. This occurs because the dedicated dataset ID for security baseline checks is not bound, and the retrieval scope covers irrelevant knowledge base content, leading to matching failure.
How to Confirm the Configuration Is Correct
- Enter a typical security baseline configuration item name to perform a single retrieval, and check if the similarity score of the returned results falls within the preset interval.
- View the source file information of the retrieval results, and confirm that they include category-specific documents such as baseline checklists and rectification suggestions.
- Check the number of retrieval results, and confirm that it matches the setting of the
limitparameter. - Compare the relevance of retrieval results with and without
usingReRankenabled, to verify that the re-ranking function is working.
The configuration values listed above are common starting points. Adjustments should be made based on specific samples and use cases for each deployment.
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.