What the data for this category looks like
This category of data originates from advertising platform backends, customer qualification records, contract systems, and publicly released regulatory rules. Daily documents include advertising review forms, qualification attachments, delivery logs, and rectification notices. Data updates are completed alongside individual advertising campaigns or natural monthly cycles. Each document typically contains fields such as campaign ID, customer name, compliance check result, and attachment hash value. Some qualification attachments are multi-page scan files, with no unified fixed field length. Total document volume fluctuates based on the scale of active advertising campaigns.
What constraints these characteristics impose on knowledge base retrieval and recall
Advertising compliance record keeping data is updated with each individual advertising campaign. The document set linked to a single campaign is relatively independent. Retrieval scope must be limited by campaign dimension to prevent irrelevant cross-campaign results from interfering with on-duty judgments. Individual document structures vary widely: there are both structured review tables and unstructured delivery logs and scan files. Multi-format retrieval logic must be supported. Data update frequency is fixed, but single campaign data volume fluctuates significantly. Sufficient index capacity must be reserved to handle peak loads. Compliance scenarios have strict requirements for result confidence. Retrieved results must be linked to their original data sources to avoid irrelevant matches impacting compliance judgments.
How to configure the settings
This configuration applies to FastGPT v4.17.0.
| Configuration item | Recommended setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the strict relevance requirements of compliance scenarios, filtering out low-relevance interfering content |
limit | Top 10 entries | Balances information density and retrieval efficiency for on-duty operations, avoiding excessive results that consume unnecessary ticket response time |
searchMode | hybrid | Adapts to the mixed format of structured review tables and unstructured delivery logs, improving retrieval rates for multiple document types |
usingReRank | Enabled | Performs secondary sorting on initial retrieval results, filtering matches with semantic deviations to meet the high confidence requirements of compliance scenarios |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval keywords to cover matching needs for synonyms and industry terms in compliance scenarios, such as expanding "qualification" to "authorization document" |
collectionIds | Bound by campaign ID | Limits retrieval scope to the knowledge base set associated with the current campaign, preventing interference from irrelevant cross-campaign data |
Three common configuration mistakes
- Phenomenon: After associating multiple knowledge bases, returned results do not allocate weights as expected, and only content from a single knowledge base is prioritized. Cause: The weight logic for multi-collection retrieval is not configured correctly, or only a single
collectionIdscollection is bound. - Phenomenon: No content can be retrieved after importing a knowledge base via mongorestore. Cause: The index rebuilding process is not triggered after import, and the vector index has not been generated. A full index rebuilding operation must be performed.
- Phenomenon: The number of retrieved results does not match expectations, and only a small number of matching items are returned. Cause: The
limitparameter is set too low, or thesimilaritythreshold is set too high, filtering out some valid compliance-related results.
How to verify successful configuration
- Navigate to the knowledge base management page, check if the
collectionIdsparameter includes all compliance document sets associated with the current campaign. - Run a single test retrieval, enter a keyword that includes the campaign ID, and verify that returned results only relate to documents from the current campaign.
- Trigger a full index rebuild, wait for the task to complete, then run a post-import retrieval test to confirm normal return of results.
- Enable retrieval logs, and check if the metadata of returned results includes fields required for compliance traceability such as document source and upload time.
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.