What the Data for This Category Looks Like
This category’s data originates from monitoring systems, ticket systems, contract documents, and customer communication logs. Updates occur monthly or quarterly. Processed daily documents include standardized SLA clauses, monthly service reports, availability statistics reports, and customer review minutes. Document fields cover service cycle, indicator thresholds, actual achieved values, abnormal event details, and contractually agreed compensation clauses. The overall structure mixes structured reports and semi-structured contract text. No uniform fixed length exists for single documents. Some contract texts contain long paragraphs with nested exclusive clauses.
Constraints on Knowledge Base Retrieval and Recall
Data for this category comes from multiple dispersed systems and has a long update cycle. Retrieval and recall must support cross-data source associated matching to avoid fragmented information. The mix of structured reports and semi-structured contract text requires retrieval strategies tailored to different text structures. Long document chunking logic must balance clause integrity and retrieval accuracy. The data includes contract-specific compensation clauses and periodic indicator thresholds. Retrieval results must strictly match the agreed fields of the current service cycle. Cross-cycle data must be prevented from interfering with subsequent SLA indicator calculation logic.
Configuration Settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.88 | Matching SLA clauses and service data has high precision requirements. Low-relevance general service records must be filtered out. |
limit | Top 6–10 entries | The number of core SLA-related data entries is moderate. Too many entries exceed the context window. Too few entries miss key indicator details. |
usingReRank | Enabled | Reranking improves the recall priority of key clauses and abnormal events for mixed structured and semi-structured data. |
datasetSearchUsingExtensionQuery | Enabled | Automatic expansion of associated contract and ticket data based on user SLA queries covers complete service cycle information. |
collectionIds | Bind to SLA-exclusive knowledge base collections | Isolate SLA-related data from other category knowledge bases to avoid irrelevant information interfering with retrieval results. |
Three Common Misconfigurations
- Symptom: Custom delimiter settings do not take effect when adjusting chunk length. Paragraph merging or splitting anomalies occur. Cause: The constraints of custom delimiters and chunk length are not matched simultaneously. Only adjusting chunk length without considering the line break format of clauses in SLA documents causes the system to ignore custom delimiters and force chunking by character count.
- Symptom: Retrieval results return a large number of irrelevant general ticket records and do not match SLA-exclusive clauses. Cause:
collectionIdsis not bound to the SLA-exclusive knowledge base collection. The retrieval scope covers the entire platform knowledge base, mixing non-target data. - Symptom: The cycle of availability statistics data recalled during SLA indicator calculation is incorrect. Cause: The retrieval scope is not limited to the current service cycle. Cross-cycle historical data is recalled, which affects indicator accuracy.
How to Verify Successful Configuration
- Upload SLA-exclusive documents and test data, run targeted retrieval, and check if returned results only include content related to the current service cycle.
- Adjust the
similarityparameter range, verify if retrieval result relevance meets business expectations, and adjust the range as needed. - After enabling
usingReRank, check if reranked results place core SLA clauses and abnormal events at the top. - Test cross-data source associated queries to confirm if contract clauses and corresponding ticket records can be recalled simultaneously.
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.