What this category of data looks like
The data for this category comes primarily from advertising platforms, internal material libraries, user interaction data, and conversion data. Updates occur on a daily or weekly basis. Daily produced documents include creative scripts, material test sheets, delivery data reports, and iteration lists. Fields covered include material type, creative combination, delivery time slot, click-through rate, conversion cost, and test conclusions. Units include metrics such as impressions, clicks, and conversions.
Constraints for Reranking and Hybrid Retrieval
Data for this category is scattered across multiple sources. Hybrid retrieval must connect advertising platform delivery data, material library files, and user interaction records at the same time. Cross-source retrieval logic must be configured. The daily or weekly update rhythm requires retrieval indexes to be synchronized regularly. This prevents expired material test results from being returned in searches. Document fields include creative combinations and delivery performance metrics. The reranking process must prioritize sorting based on business-related fields such as conversion cost and click-through rate. It uses text similarity to help optimize ranking results. The high timeliness requirement of creative testing limits the number of valid results returned per single search. It also requires filtering material-related content that carries violation risks to avoid harming delivery performance.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
usingReRank | Enabled | Feed-based creative testing requires sorting based on delivery performance fields. Reranking improves result relevance and aligns with the built-in reranking model invocation logic of FastGPT v4.17.0 |
similarity | 0.65–0.75 | Filters low-relevance creative scripts and test sheets to avoid interfering with valid retrieval results, and aligns with the text characteristics of business documents |
limit | Top 8–12 entries | The number of valid creative test combinations is limited. Excessive results increase screening costs and meet the rapid positioning needs of operations on-duty |
searchMode | Hybrid Retrieval | Must simultaneously retrieve material libraries, delivery data, and user interaction data to cover all business documents |
| Reranking model matching fields | Click-through rate, conversion cost | Matches the business characteristics of image, video, and copy combinations. Prioritizes sorting based on core business metrics, which aligns with the core comparison dimensions of feed-based creative testing |
Three Common Misconfigurations
- Symptom: No retrieval results appear after enabling the reranking model in the application, while results are normal when testing the knowledge base separately. Cause: Cross-source retrieval association fields are not configured. Only single-source data is loaded during knowledge base testing, and advertising platform and material library data are not correctly linked when the application is invoked.
- Symptom: Results returned after reranking include content related to non-compliant materials. Cause: Violation keyword filtering logic is not added to the reranking process. This causes low-quality or non-compliant creative combinations to be prioritized in rankings, which may lead to delivery risks.
- Symptom: The number of single retrieval results exceeds expectations, making it impossible to quickly locate valid test conclusions during operations on-duty. Cause: The value range of the
limitparameter is not restricted, resulting in redundant invalid results being returned.
How to Verify Proper Configuration
- Navigate to the FastGPT application configuration page and confirm that the
usingReRanktoggle is enabled. - Simulate an operations on-duty scenario, enter the keyword "last week's high-conversion creative combinations", and verify that the retrieval results include material test sheets and delivery data report content.
- Adjust the
similarityparameter value and verify whether the relevance of retrieval results meets business expectations. - View the retrieval log and confirm that
searchModeis set to Hybrid Retrieval mode, and that multi-source data interfaces are called simultaneously.
The values provided here 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.