What the Data for This Category Looks Like
The data for this category primarily comes from advertising platforms, customer relationship management systems, attribution SDKs, and online form systems. Core documents for daily maintenance include attribution rule configuration tables, lead distribution tables, daily channel conversion reports, and abnormal lead checklists. Data update frequencies are mostly daily or weekly. Fields include channel unique identifiers, lead source parameters, conversion timestamps, sales attribution accounts, and more. Units include timestamps, count units, currency units, and others. Data formats vary across different sources.
What Constraints Do These Characteristics Impose on Reranking and Hybrid Retrieval?
The multi-source, heterogeneous data characteristics of this category require the hybrid retrieval link to be compatible with field mapping rules for different data sources, and to set dedicated retrieval fields for different document types such as attribution rules and lead distribution tables. The daily or weekly update frequency requires that retrieval prioritizes the latest channel reports and abnormal checklists, so a weight adjustment mechanism based on update time must be configured. Structured daily documents require the reranking link to perform precise matching on key fields such as lead IDs and channel parameters, to avoid interference from non-key fields on retrieval results. The accuracy of sales attribution-related data directly affects retrieval priority, so a weight factor for attribution credibility must be added to the reranking link to ensure that ticket Q&A results align with business logic.
How to Configure the Settings
The following configurations are based on FastGPT v4.17.0, adapted to the business requirements of the channel attribution and lead distribution scenario:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Reranking Toggle | Enabled | This category requires precise matching of lead attribution and channel parameters. Enabling reranking filters low-relevance retrieval results and improves ticket Q&A accuracy |
| Similarity Threshold | 0.72–0.85 | Most data for this category consists of structured fields. This range balances recall rate and ranking accuracy, and can be adjusted based on actual measurements |
| Single Recall Limit | Top 8–12 entries | The scale of daily documents for this category is moderate. This range covers all key leads and channel data, avoiding interference from redundant information |
| Retrieval Mode | Hybrid Retrieval | Data for this category comes from multiple source systems. Hybrid retrieval can match both structured fields and unstructured text, adapting to different document types |
| Reranking Model Selection | Adapt to Data Source | Field characteristics differ between advertising platform data and CRM data. Configure corresponding reranking models for different data sources to improve matching effectiveness |
| Context Window | 1000–1500 token | Ticket Q&A for this category requires complete attribution rules and lead information. This window covers core context content and avoids information truncation |
Three Common Configuration Mistakes
- Symptom: The rerank marker field in retrieval results returns false, and the sorted results after reranking are not displayed in the interface. Cause: The reranking toggle is not enabled, or the reranking model configuration is not bound to the corresponding data source.
- Symptom: When adding a reranking model with the same name, the existing model configuration with the same name is automatically overwritten. Cause: The system only supports a single reranking model configuration per name. Configure model names with unique identifiers for different data sources.
- Symptom: Calling the reranking model returns empty content with no valid sorting results. Cause: The input parameter format of the reranking model does not match the data source fields, or the model call timed out without returning a result.
How to Confirm the Configuration Is Complete
- Navigate to the FastGPT retrieval test page, enter channel parameter verification questions, and check whether the retrieval results include the latest channel reports and attribution rule documents.
- View the system’s reranking model call logs to confirm that reranking processes are triggered for each retrieval, and that the returned result sorting aligns with business priorities.
- Check the binding configuration for multiple data sources, confirm that different types of documents such as lead distribution tables and abnormal checklists are bound to corresponding reranking models.
- Simulate an abnormal lead query scenario, confirm that retrieval results prioritize displaying high-credibility sales attribution and channel data.
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.