What data for this category includes
Daily documents for this category include training slides, drill plans, phishing email templates, and drill result reports. Data sources cover employee contact lists, email gateways, internal training platforms, and user click statistics. Data update frequency is typically quarterly or annual. Most document structures include scenario descriptions, operation steps, statistical fields, and result review modules. Fields are mostly linked to personnel groups, drill batches, and security behavior data. No high-frequency real-time updated business data exists for this category.
What constraints these characteristics impose on knowledge base retrieval and recall
First, data update frequency is low, so high-frequency real-time recall logic is not required. Retrieval results must still highly match current drill scenarios. Second, document types include professional text and template content. Precise recall of multi-format content must be supported to avoid missing key details. Third, data sources are linked to multiple business systems. Filtering the recall scope by collection dimension must be supported, to ensure retrieved content only relates to security training business. Fourth, business error consequences directly tie to security risks. The number of redundant recall results must be limited, to avoid irrelevant content interfering with security decision-making and review work.
How to set configurations
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Similarity Threshold | 0.75–0.85 | Matches the semantic similarity between phishing email templates and drill scenarios. Avoids too low a value that lets irrelevant content mix in, and too high a value that misses applicable scenarios. |
| Single Search Limit | Top 5–10 entries | Controls the number of recall results per search, reduces information redundancy in security training scenarios, and adapts to the reading rhythm of review and plan design. |
| Rerank Function | Enabled | Performs secondary sorting on recall results, improves matching accuracy for professional content such as phishing email templates and drill plans. |
| Collection Filtering | Specify 2–3 security training-specific collections | Limits the retrieval scope to security training-specific datasets, excludes irrelevant content from other business collections, and complies with data source isolation requirements. |
| Query Expansion | Disabled | This category of data has a low update frequency, so no additional query semantic expansion is needed. Avoids introducing irrelevant general security content. |
| Search Mode | Semantic Recall | Adapts to the semantic matching needs of professional documents, replaces the ambiguity of keyword recall, and improves the matching accuracy of templates and plans. |
Three common misconfigurations
- Retrieval results include documents from multiple business collections, and cannot filter out security training-specific content. The cause is failure to configure the collection filtering parameter, and failure to limit the retrieval scope to security training-specific datasets.
- A configuration error prompt pops up when clicking the knowledge base creation function. The cause is failure to correctly associate collection permissions for data sources such as employee contact lists and email gateways, or the similarity threshold value is outside the reasonable range.
- Recalled phishing email template content is incomplete, only returning fragmentary text. The cause is failure to enable the document content full reading configuration, or the single search limit value is set too low, causing content to be truncated.
How to confirm the configuration is properly set
- Run a retrieval test, enter keywords related to phishing email templates, confirm that only documents within the security training-specific collections are returned.
- Adjust the similarity threshold range, verify that the semantic matching degree of recall results meets the needs of the current drill scenario.
- After enabling the rerank function, compare the correlation difference between the results before and after sorting, confirm that the sorting of professional content is more reasonable.
- Check the single search limit value, confirm that the number of single recall results adapts to the reading rhythm of business review and plan design.
The values provided are common starting points and should be evaluated against samples relevant to the specific 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.