What data belongs to this category
Core data for this category comes from employee directories, email gateways, internal training platforms, and user click statistics systems. Daily produced documents include standardized training courseware, phishing drill execution plans, bulk phishing email templates, and drill result review reports. Data update cycles are mostly quarterly or annual. Most documents are report or template files with fixed fields. Some template documents have fixed-format parameter placeholders for personalized configuration across different drill scenarios.
What constraints do these characteristics impose on knowledge base retrieval and recall
Data sources are scattered and include cross-system employee attributes and email behavior data. Knowledge base recall must support cross-dataset associated queries to avoid missing information from only searching single document types. Document types include templates and review reports. Some templates have fixed placeholders. Retrieval logic must be configured to ignore matching meaningless placeholders to avoid filtering valid template content. Quarterly update frequency means high-frequency incremental synchronization is not needed, but precise filtering by drill scenario and training cycle must be supported. Drill audience screening relies on employee directory fields. Retrieval must associate employee attributes such as department and position to narrow recall scope and improve query efficiency.
The following configurations are based on FastGPT v4.17.0.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches text similarity for phishing email templates, avoids mismatched placeholders or low-relevance training courseware |
limit | Top 8 entries | Balances recall volume for drill plans and review reports, covers query needs for single operational duty shifts |
searchMode | hybrid | Combines keyword and semantic retrieval, adapts to fixed fields and semantic queries for template documents |
usingReRank | Enabled | Improves sorting accuracy for phishing email templates and drill result reports, filters irrelevant training courseware |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands associated queries for employee directories and click statistics, supports retrieval needs for drill audience screening |
collectionIds | Bind corresponding datasets by scenario | Separates datasets for training courseware, drill plans, and email templates, avoids interference from non-category documents during recall |
Three common mistakes
- Insufficient retrieval results with large numbers of irrelevant documents, and displayed results have low matching with query topics. Cause: Failed to limit retrieval scope using
collectionIds, recalled documents from non-security training datasets of other business lines. - When querying for phishing email template optimization suggestions, the system cannot associate employee directory data or complete retrieval related to audience screening. Cause:
datasetSearchUsingExtensionQueryis not enabled, and cross-dataset associated query logic is not configured. - After uploading Excel-format drill result reports, retrieval performance is poor and structured data in tables cannot be extracted. Cause: Reasonable chunking rules for table structured data are not configured. Directly uploading entire tables leads to unclear semantic segmentation and insufficient vector matching accuracy.
How to confirm configurations are set correctly
- Execute a query targeting phishing email templates, check if returned results only include preset security training category dataset documents.
- Trigger a query containing employee position information, confirm if recalled results associate drill plans from the corresponding department.
- Adjust the
similaritythreshold, verify if the relevance of recalled results meets matching requirements for the business scenario. - Upload a test document containing images, confirm if text content within images is correctly extracted during retrieval.
The values provided here are common starting points, and should be evaluated 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.