What Data Looks Like for This Category
Data for landing page conversion optimization scenarios mainly comes from landing page access logs, heatmaps, form data, and ad source parameters. Daily processed documents include landing page plans, event tracking requirement documents, A/B test reports, and form field lists. Data update frequency is weekly or per individual marketing campaign milestones. Fields include page path, dwell time, form submission status, ad channel identifiers, and more. Analysis dimensions must be adjusted based on the specific needs of a single campaign, and there is no unified fixed statistical caliber.
What Constraints Do These Characteristics Impose on the Reranking and Hybrid Retrieval Link
Landing page conversion optimization data covers user behavior and advertising delivery workflows. Relevant documents must be matched to specific landing page scenarios mentioned in user inquiries, so reranking and hybrid retrieval must prioritize associating exclusive documents for the current campaign. Data update frequency fluctuates with campaign milestones, requiring retrieval logic to quickly adapt to newly uploaded event tracking, test report, and other documents. Most documents are scenario-based operational content. Irrelevant general knowledge base content must be avoided. At the same time, recall coverage and result accuracy must be balanced to prevent excessive redundant information from interfering with the efficiency of pre-sales Q&A and technical support.
How to Configure
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
usingReRank | Enabled | Most documents for landing page conversion optimization are scenario-based operational content. Reranking filters low-relevance general search results, improving the accuracy of pre-sales Q&A and technical support |
limit | Top 6-8 entries | The volume of landing page documents related to a single campaign is usually limited. Limiting the number of recalled entries avoids information overload, meeting the concise needs of pre-sales inquiries |
similarity | 0.72-0.85 | Balance recall coverage of exclusive landing page documents and filtering of irrelevant content, avoiding omission of professional content related to event tracking and A/B testing |
searchMode | hybrid | Combine semantic retrieval and landing page behavior data retrieval to match specific campaign scenarios mentioned in user inquiries, adapting to the scenario-based needs of pre-sales Q&A |
Three Common Misconfigurations
- Phenomenon:
usingReRankis enabled, and reranking is configured during the knowledge base indexing phase, but no reranked results appear in online recall tests. Cause: The reranking switch was not enabled synchronously during the retrieval phase, or the API key for the reranking model was configured incorrectly. - Phenomenon: The reranking model call returns the
Invalid URL (POST /v1/rerank)error. Cause: The API endpoint path for the reranking model was not configured correctly, or the third-party reranking service did not start normally. - Phenomenon: The reranking model fails to load properly after deployment. Cause: The format of the deployed reranking model does not match the format supported by FastGPT v4.17.0, or the environment packages required for model operation are missing.
How to Confirm the Configuration is Successfully Applied
- Navigate to the retrieval test page of the FastGPT knowledge base, enter an inquiry that includes landing page form fields or event tracking requirements, and check whether the recalled results have their sorting order adjusted according to the reranking logic.
- Check the reranking model call logs to confirm there are no records of connection timeouts or parameter errors.
- Compare the number of recalled results and sorting differences between when the reranking function is enabled and when it is disabled, to verify that the reranking logic has taken effect.
- Upload a new landing page campaign document, confirm that the retrieval results can cover this new document and the sorting meets expectations.
The values provided are common starting points and should be evaluated using samples specific to the respective deployment.
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.