What the data for this category looks like
Core data for this category comes from ad delivery platforms, asset libraries, user interaction data, and conversion data. Daily produced documents include creative scripts, asset test sheets, delivery data reports, and iteration checklists. Data update frequency is typically daily or weekly. Most document structures include fields such as creative variables, test batches, delivery metrics, and iteration notes. There are no unified mandatory units, and values must adapt to specific business documents. Core business revolves around combined testing of images, videos, and copy. Data relevance relies on the corresponding relationship between test variables and delivery effects.
Constraints imposed on knowledge base retrieval and recall
Decentralized data sources require integration of multi-source heterogeneous content. Differences in update frequencies require prioritizing the recall of the latest delivery and test data during retrieval. Diverse document types require adaptation to text splitting and indexing for different formats such as scripts, tables, and reports. The core of creative testing is comparing combined effects, so retrieval must support filtering by dimensions such as variables and batches. The time-consuming step of summarizing test conclusions requires recall results to quickly associate with historical test data, avoiding repeated analysis. Material compliance risks require retrieval to synchronously link platform audit rule documents to ensure content compliance.
How to set configurations
The following configurations are formulated for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches retrieval terms related to creatives, filters low-relevance non-ad test content |
limit | Top 10 entries | Covers common creative variable combinations and historical test data, avoids excessive results interfering with summarization |
searchMode | Semantic retrieval | Adapts to semantic matching requirements for unstructured text such as creative scripts and test reports |
usingReRank | Enabled | Re-ranks initial recall results, prioritizes displaying assets and data most relevant to current testing goals |
collectionIds | Specified by business collection | Limits retrieval scope to the knowledge base collection exclusive to ad creative testing, prevents mixing data from other categories |
datasetSearchUsingExtensionQuery | Enabled | Expands retrieval terms to cover common synonyms and variant expressions in creative testing, such as "asset fatigue" and "asset aging" |
Three common misconfigurations
- Phenomenon: Retrieval results include non-creative test documents, and filtering by business collection is not possible. Cause: The
collectionIdsparameter is not configured, and the retrieval scope is not limited to the exclusive knowledge base collection for ad creative testing. - Phenomenon: In FastGPT v4.8.10, retrieval only displays 1 original query, and does not show expanded related questions. Cause: The
datasetSearchUsingExtensionQueryconfiguration is not enabled, and the retrieval term expansion function is not activated. - Phenomenon: Custom indexes are not created as needed, making it impossible to quickly locate documents related to creative testing. Cause: The role of indexes for isolating data by business domain is not clarified, and exclusive indexes are not created for the creative testing scenario.
How to verify successful configuration
- A retrieval term related to creative testing is manually input, and retrieval results are checked to confirm they only include ad creative test documents, verifying the
collectionIdsconfiguration is effective. - The number of returned retrieval results is checked, and the quantity is confirmed to match the
limitparameter setting, verifying the retrieval upper limit configuration is correct. - The sorting of retrieval results with
usingReRankenabled and disabled is compared, confirming the re-ranking function works properly. - Expanded retrieval terms are checked to confirm they cover synonyms and variant expressions, verifying the
datasetSearchUsingExtensionQueryconfiguration is enabled.
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.