What the data for this category looks like
This category’s data primarily comes from customer ITSM systems, operations monitoring platforms, asset inventory databases, and on-site duty records. Daily produced documents include on-site daily reports, service desk tickets, inspection reports, and SLA monthly reports. Document fields include ticket numbers, inspection items, SLA compliance status, duty periods, and more. Data updates follow a mostly daily schedule: on-site daily reports and ticket records update each day. SLA monthly reports update via monthly aggregation. The data structure centers on the full on-site service delivery workflow, linking customer-side IT assets and service request information. Field formats vary across different data sources.
What constraints these characteristics impose on the reranking and hybrid retrieval workflow
Dispersed data sources linked to multiple systems require hybrid retrieval to integrate tickets, inspection reports and other content across sources with different formats. Clear field mapping rules must be defined for each data source to avoid field mismatch issues. Differences in update frequencies require distinguishing real-time data and historical monthly reports during retrieval. Prioritize daily tickets and inspection records to avoid mixing information of different timeliness that could reduce accuracy. Most on-site service documents are process records with strong field correlation. The reranking workflow must prioritize core business fields such as ticket priority and SLA timeliness. It must also adhere to business regulations, which constrain the filtering dimensions and display order of retrieval results.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Search Mode | Hybrid Retrieval Mode | Adapt to multi-source heterogeneous on-site service documents, balance semantic matching and keyword recall, cover different types of data such as tickets and inspection reports |
| Similarity Threshold | 0.72–0.80 | Filter low-relevance duty records and non-core inspection items, retain content strongly linked to customer technical issues, calibrated via actual testing |
| Single Search Limit | Top 12 entries | Control the number of initially recalled documents, match the daily update volume of on-site service data, avoid context overload |
| Reranking Toggle | Enabled | Address the field correlation of multi-source data, optimize the ranking of core fields such as ticket priority and SLA timeliness via reranking, improve retrieval result matching accuracy |
| Reranking Recall Limit | Top 6 entries | Narrow the scope of results after reranking, focus on the most relevant tickets and inspection records, meet the information density requirements of pre-sales and technical support |
| FastGPT Version | v4.17.0 | Adapt to the reranking and hybrid retrieval function logic of this version, ensure configuration items take effect |
Three Common Mistakes
- Phenomenon: Hybrid retrieval returns a large number of non-target documents, such as unrelated duty records. Cause: No filtering rules are configured for the fields of on-site service documents, and the retrieval scope is not limited to ticket and inspection report data.
- Phenomenon: The reranking toggle displays as enabled in the interface, but retrieval results do not change. Cause: The selected retrieval mode does not support the reranking function, or reranking trigger conditions are not bound to retrieval requests.
- Phenomenon: Multiple variables following an HTTP request in a workflow cannot be bound to a specified knowledge base. Cause: No corresponding data source filtering rules are set for different variables in the reranking configuration, and classification fields of on-site service documents are not matched.
How to Confirm Configurations Are Set Correctly
- View retrieval logs to confirm that returned document types include target category documents such as on-site daily reports and service desk tickets. Verify that document sources match the configured data sources.
- Test enabling and disabling the reranking function, compare the sorting logic of retrieval results, and confirm that reranked results prioritize high-priority tickets and compliant inspection records.
- Adjust the similarity threshold and single search limit, observe changes in the number and relevance of retrieval results, and confirm that configuration parameters affect retrieval results as expected.
- Input keywords from non-target categories to verify that filtering rules take effect. Confirm that retrieval results do not return unrelated asset inventory or duty records.
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.