What the data for this category looks like
Daily documents for landing page conversion optimization include four core types: landing page plans, point burying requirements, AB test reports, and form field lists. Data sources mainly include landing page access logs, heatmap data, form submission data, and ad source parameters. The data update frequency is weekly or within a single campaign cycle. Fields include page access path, dwell time, form submission status, ad channel identifier, and others. The content structure adjusts flexibly based on specific campaigns, with no fixed unified format.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Multi-source and scattered data sources require retrieval to cover the four core document types: landing page plans, point burying requirements, AB test reports, and form field lists, to avoid missing key business information with a single retrieval mode. The weekly or single campaign cycle update rhythm requires the knowledge base synchronization frequency to match the business rhythm, to avoid recalling outdated old data. The flexible document structure with no fixed unified format requires retrieval to balance keyword matching and semantic understanding, adapting to different types of business questions. Dispersed fields that adjust with campaigns require retrieval configuration to support associating knowledge bases by business grouping, to avoid interference from cross-category data.
Configuration Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Match business keywords related to landing pages, filter low-correlation general operation and maintenance documents |
limit | Top 8 entries | Adapt to single-round on-duty Q&A scenarios, control context length to not exceed the model's processing limit |
searchMode | hybrid | Combine keyword and semantic retrieval, cover different types of documents such as point burying requirements and AB test reports |
usingReRank | Enabled | Perform secondary sorting on recalled results, prioritize matching core business content such as form fields and ad attribution |
datasetSearchUsingExtensionQuery | Enabled | Optimize extended expressions of user questions, match detailed content in landing page plans |
collectionIds | Configured by business grouping | Only associate knowledge base collections exclusive to landing page conversion optimization, avoid interference from cross-category data |
Three Common Misconfigurations
- Phenomenon: When calling the
/api/v1/chat/completionsinterface, the corresponding knowledge base content is not recalled after specifying the appId. Reason: The target knowledge base collection is not associated via thecollectionIdsconfiguration, or the configured collection ID does not match the collection where the document was actually uploaded. - Phenomenon: The retrieval results contain a large number of irrelevant general operation and maintenance documents, and landing page-related content is not hit. Reason: The
similaritythreshold is set too low, recalling non-core documents outside the business scope. - Phenomenon: The context returned in a single round of Q&A is too long, causing an error exceeding the model's processing limit. Reason: The
limitparameter is set too large, failing to control the number of recalled results.
How to Verify Correct Configuration
- Call the application-associated knowledge base verification interface provided by FastGPT, check whether the knowledge base collection ID bound under the specified appId matches the configured
collectionIds. - Enter business keywords such as landing page path, form field, etc., to verify whether the retrieval results preferentially return landing page plans and AB test report documents.
- Adjust the
similaritythreshold, observe the correlation changes of the recalled results, and confirm that the threshold meets business requirements. - After enabling
usingReRank, compare the result ranking before and after reranking to confirm that the reranking logic takes effect.
The values provided are common starting points and should be evaluated 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.