What data for this category looks like
Data for this category primarily comes from advertising platforms, landing pages, product qualifications, and advertising law clauses. Update frequency is daily or real-time. Daily processed documents include ad creatives, landing page screenshots, review opinions, and modification records. Core fields include creative unique identifiers, violation type tags, review status, and modification timestamps. The length of individual creative documents varies widely, ranging from copy assets of tens of characters to landing page screenshot files of hundreds of pixels.
Constraints Imposed on Reranking and Hybrid Retrieval by These Characteristics
The multi-source heterogeneous nature of ad creatives requires hybrid retrieval to support both text and image metadata. A multimodal retrieval configuration is therefore needed. The daily or real-time update frequency of creatives means the retrieval pipeline must support real-time synchronized index updates. This prevents expired assets from being recalled. The wide variance in individual creative document lengths requires the reranking step to accommodate inputs of different lengths. This avoids retrieval failure caused by context overflow. Compliance documents associated with reviews must strictly match the latest versions. Reranking results must prioritize content from currently active advertising law clauses. This ensures recalled content meets compliance requirements.
How to Set Configurations
These configuration items apply to FastGPT v4.17.0.
| Configuration Item | Recommended Values | Rationale |
|---|---|---|
| Similarity threshold | 0.75–0.85 | Aligns with the strictness of ad creative compliance retrieval, balances recall precision and coverage |
| Maximum single search results | Top 8–12 results | Controls the number of results returned per search to avoid computational overload in the reranking step |
| Retrieval mode | Hybrid retrieval | Supports multi-source data such as ad creative text and landing page metadata |
| Enable reranking | Enabled | Optimizes the relevance ranking of recall results to meet the high compliance requirements of ad creative review |
| Reranking recall limit | Top 5–7 results | Focuses on high-priority compliance matching results, reduces unnecessary validation |
Three Common Misconfigurations
- Reranking results do not appear during online recall testing. This occurs when reranking-associated configuration is not enabled during the knowledge base indexing step, or when the reranking function is not enabled during the retrieval step.
- A reranking model connection error displays Invalid URL (POST /v1/rerank). This happens if the reranking model’s API address is configured incorrectly, or if the third-party API forwarding service’s reranking interface path does not match the format required by FastGPT.
- Local reranking model requests fail. This is caused by the local model’s port not being open, or the configured reranking model parameters being incompatible with the model formats supported by FastGPT.
How to Confirm Configurations Are Correct
- Navigate to the FastGPT retrieval configuration page. Confirm the enable reranking toggle is switched on.
- Submit a test query related to ad creatives. Check if the retrieval result list includes reranking score fields.
- Review the reranking model’s API configuration information. Confirm that parameters such as address and API key match the deployed model service.
- Trigger a full knowledge base index. Confirm that the reranking module does not throw exception logs during the indexing process.
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.