What the data for this category looks like
Data sources for this category include advertising platform APIs, user profile tag libraries, budget scheduling tables, and attribution tool logs. Update frequency follows two modes: daily sync and single activity trigger. Daily documents include campaign master tables, targeting setting rules, budget allocation plans, and conversion tracking configuration tables. Fields include targeted audience package ID, budget amount, conversion event type, delivery time slot parameters, and more. Document structure is divided into three modules based on the ad launch workflow: pre-launch configuration, execution parameters, and effect tracking.
What constraints these characteristics impose on knowledge base retrieval and recall
Differing update rhythms across multiple data sources require the retrieval link to support filtering latest configurations by timestamp. This prevents referencing expired budget or targeting rules. Documents that mix structured fields and text descriptions require retrieval to balance semantic matching and precise keyword matching. Independent documents for different modules require retrieval to allow limiting recall scope by module. This avoids interference from documents of other categories. Data scale grows with the number of campaign activities. The recall upper limit must match single-round context carrying capacity to prevent exceeding model input limits.
How to set configurations
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.75-0.85 | Campaign configuration fields have clear semantics. This range filters irrelevant results while retaining valid targeting rules |
limit | Top 6-10 entries | Individual campaign configuration entries are relatively few. Excessive results will exceed the conversation context window |
searchMode | vector+keyword hybrid mode | Data includes structured ID fields and text descriptions. Hybrid retrieval balances semantic association and precise keyword matching |
usingReRank | Enabled | Targeting rules for campaigns have associated logic. Reranking optimizes result priority ordering |
datasetSearchUsingExtensionQuery | Enabled | Campaign configuration relies on associated documents. Extended queries can supplement context information |
collectionIds | Bind dedicated knowledge base collections for campaigns | Avoid mixing irrelevant documents from other categories, ensuring recall results accurately match the current task |
Three common configuration errors
- Symptom: Knowledge base search works correctly in workflow debugging, but no knowledge base reference results are returned in AI conversations. Cause: Knowledge base recall binding is not configured during the conversation phase, or the bound knowledge base collection does not match the one specified in the workflow.
- Symptom: Unable to limit the number of knowledge base references via parameters, only filtering by score threshold. Cause: The
limitparameter is not used correctly to control the number of returned entries, andsimilarityis mistakenly used as the sole method for limiting entry count. - Symptom: After uploading a text file with embedded images, retrievable image-associated URLs cannot be generated. Cause: Image OCR parsing configuration is not enabled, or the uploaded image format does not meet system support requirements.
How to verify successful configuration
- Enter the FastGPT v4.17.0 knowledge base management page, check whether the bound
collectionIdsincludes document collections related to campaigns. - Initiate a test conversation, enter specific questions about targeting settings or budget allocation, and confirm returned results include knowledge base configuration entries.
- Adjust the value range of the
similarityparameter, observe changes in retrieval result relevance, and confirm parameter configuration has taken effect. - Enable conversation debugging mode, view the recall log, and confirm
usingReRankanddatasetSearchUsingExtensionQueryhave executed normally.
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.