What the Data for This Category Looks Like
This category of data primarily originates from advertising platform backends, customer qualification libraries, contract systems, and publicly available regulatory rules. The core documents processed daily fall into four categories: advertising review forms, qualification attachments, delivery records, and rectification notices. Data updates trigger upon completion of a single advertising campaign or on a monthly cycle, with no fixed high-frequency update rhythm. Data fields include activity ID, delivery time period, compliance review status, qualification number, and other relevant fields. There are no unified fixed units; values shift dynamically based on document type.
Constraints for Knowledge Base Retrieval and Recall
Decentralized data sources require retrieval to cover multiple independent datasets, and support cross-dataset recall configuration. Single campaign or monthly update rhythms require retrieval to support incremental synchronization mechanisms, to avoid redundant full re-scanning of data. Structural differences across multiple document types require retrieval to match different field weights. For example, qualification attachments should prioritize matching number fields, while delivery records should match activity ID. Regulatory accountability risks require recall results to retain complete metadata, to ensure traceability back to the original activity or qualification.
Configuration Settings
These configurations are written for FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Balances precise matching and recall coverage for compliance documents, avoids missing critical qualifications or delivery records |
limit | Top 10 entries | Single recall results need to cover multiple document types; 10 entries balances completeness of compliance evidence and context length |
searchMode | keyword + vector | Adapts to multiple document types. Keyword matching quickly locates qualification numbers, while vector matching suits unstructured rectification notices |
usingReRank | Enabled | Compliance documents have high accuracy requirements; re-ranking filters low-relevance results to ensure evidence validity during regulatory inspections |
datasetSearchUsingExtensionQuery | Enabled | Supplements context related to advertising activities, improves matching accuracy for delivery records and review forms |
collectionIds | Configured by grouping compliance datasets | Only recalls datasets related to advertising compliance, avoids interference from unrelated documents in retrieval results |
Three Common Misconfigurations
- Issue: When uploading a compliance dataset in CSV format, using a custom delimiter fails to correctly split file lines. Cause: The custom delimiter option was not selected in the dataset upload interface, or the delimiter input contains invisible whitespace characters.
- Issue: After configuring authentication for the knowledge base search node, authentication verification does not take effect during calls. Cause: Authentication parameters were not bound to the permission configuration item of the corresponding search node, or the node permission scope does not cover the target compliance dataset.
- Issue: When selecting a variable reference for the knowledge base during API calls, no optional values are available. Cause: No input variable for knowledge base selection was configured in the workflow, or the variable was not bound to the knowledge base parameter of the search node.
How to Verify Successful Configuration
- Navigate to the dataset management interface, and confirm that all configured compliance datasets are associated with the current search node.
- Trigger a test retrieval, and check whether the returned result metadata includes core compliance document fields such as activity ID and qualification number.
- Call the API interface, compare the number of results and similarity distribution between the online conversation and API returns to confirm that configuration parameters are consistent.
- Simulate a compliance scenario with a keyword input, check whether corresponding qualification attachments or delivery records are recalled to verify that the retrieval logic is active.
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.