What This Category of Data Looks Like
This category of data primarily originates from advertising platform backends, customer qualification documents, contract systems, and publicly available regulatory rules. Routinely processed documents include advertising review forms, qualification attachments, delivery records, and rectification notices. Most individual documents contain fields such as activity identifiers, qualification numbers, delivery periods, and rectification requirements. Data updates are triggered upon completion of a single advertising campaign or synchronized on a monthly basis. Each data entry is associated with full-process compliance evidence for its corresponding advertising campaign, with no redundant cross-campaign associations.
Constraints for Knowledge Base Retrieval and Recall
Advertising compliance record-keeping data is linked to individual advertising campaigns. Retrieval must accurately match activity identifiers to avoid recalling irrelevant compliance evidence across campaigns. Data updates occur per campaign or monthly, so the system must support quick filtering of historical data by time range. Individual documents hold multiple types of compliance materials, so retrieval must differentiate recall weights for qualification-related and record-related documents to ensure core compliance evidence is returned first. Recall results must also be complete, to avoid missing valid archived materials during regulatory inspections and compromising compliance verification.
Configuration Settings
The following configurations are adapted for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Semantic similarity threshold for matching advertising compliance documents, filters low-relevance cross-campaign compliance materials |
limit | Top 10 entries | Controls the number of single-recall results, adapted to the scale of compliance documents associated with a single advertising campaign |
searchMode | vector_only | Prioritizes semantic matching for compliance documents, avoids generating irrelevant general Q&A content |
usingReRank | Enabled | Re-ranks recall results by compliance relevance, ensures core materials such as qualification attachments and delivery records are displayed first |
datasetSearchUsingExtensionQuery | Disabled | Does not generate additional query terms, avoids deviating from the precise matching scope of compliance documents |
collectionIds | Specified by compliance document collection | Limits the retrieval scope to only the exclusive knowledge base collection for advertising compliance record-keeping, excludes data from other categories |
Three Common Misconfigurations
- Phenomenon: Retrieval results include unuploaded AI-generated content, with no matching compliance document content. Cause:
datasetSearchUsingExtensionQueryis enabled, and the system automatically generates additional query terms, causing the recall scope to exceed the compliance knowledge base. - Phenomenon: The interface prompts that the model does not exist or is inaccessible. Cause: The exclusive available model is not bound to the advertising compliance record-keeping knowledge base, or an unauthorized model version is called.
- Phenomenon: Images are missing in uploaded docx documents. Cause: The document image extraction configuration is not enabled, or the embedded image format of the original document was not retained during upload, causing qualification attachment images in compliance materials to fail to load properly.
How to Verify Successful Configuration
- Upload a test advertising review form, retrieve activity identifier keywords, and verify that the returned results only include compliance documents for that campaign, with no irrelevant content.
- Adjust the
similaritythreshold, verify changes in recall result relevance, and ensure core compliance materials are returned first. - View the
collectionIdsconfiguration of the knowledge base collection, confirm that only the exclusive document collection for advertising compliance record-keeping is included, with no cross-category data mixed in. - Upload a docx qualification attachment with images, verify that images are displayed normally in the retrieval result preview.
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.