What this category of data looks like
This category of data primarily comes from cloud billing systems, tag data, order systems, and resource monitoring tools. Documents processed daily include billing details, cost allocation tables, resource utilization reports, and optimization suggestions. Data fields cover resource identifiers, expense details, tag attribution, allocation rules, and more. Units include monetary amounts, resource usage quantities, time cycles, and others. Data update frequency is typically daily or monthly. The update rhythm of different reports aligns with business reconciliation and analysis cycles.
What constraints do these characteristics impose on knowledge base retrieval and recall?
Decentralized data sources lead to differences in field formats across different documents. Unified field mapping rules must be matched during retrieval to avoid misaligned recalled content. Differences in update frequencies require prioritizing the latest daily billing data during retrieval, while also supporting monthly historical allocation reports. Diverse document types include both structured detailed tables and unstructured optimization suggestions, requiring adaptation to text splitting and recall logic for different formats. The high time-consuming requirements of abnormal expense troubleshooting and cost allocation explanations require that retrieved and recalled results accurately match the query's business scenario, reducing interference from invalid content.
How to configure settings
This page provides configuration instructions based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the field accuracy of cloud billing data, avoids recalling irrelevant resource monitoring data while ensuring sufficient recall coverage |
limit | Top 6 entries | Balances the content volume of a single retrieval and query response speed, adapts to the real-time Q&A scenario of operations duty, and avoids returning excessive redundant information |
searchMode | Hybrid retrieval | Balances the recall effect of structured billing details and unstructured optimization suggestions, covering different types of business documents |
usingReRank | Enabled | Ranks recalled multi-type documents, prioritizing content directly related to cost allocation and abnormal troubleshooting |
datasetSearchUsingExtensionQuery | Disabled | Query terms for cloud billing data typically point to clear business scenarios, so no additional query expansion is needed, avoiding introduction of irrelevant information |
Three common configuration errors
- Phenomenon: Hybrid retrieval response time exceeds 10 seconds, with initial vector retrieval response taking approximately 8 seconds. Cause: Hybrid retrieval must call recall logic for both structured and unstructured data simultaneously. The high complexity of field associations in cloud billing data increases retrieval latency.
- Phenomenon: After uploading a billing-related PDF document, no parsing completion prompt appears in the interface, and search test results are empty. Cause: Document parsing trigger rules are not configured correctly, or the uploaded PDF contains encrypted content leading to parsing failure and no valid vector data generated.
- Phenomenon: The Q&A system does not recall knowledge base content, directly generating generic responses with content deviations. Cause: The
similaritythreshold of the retrieval configuration is set too high, making it impossible to recall matching billing details or allocation table content, or the correctcollectionIdsfor the target knowledge base collection are not specified.
How to confirm proper configuration
- Run a single retrieval test, verify that the recalled result fields include core content related to cloud billing and cost allocation, and adjust configuration items to the range that meets business requirements.
- Check retrieval logs to confirm that response time meets the real-time interaction requirements of operations duty, and adjust the
limitandusingReRankconfigurations based on actual scenarios. - Upload a test billing detail document, verify that a parsing completion prompt is generated, and search tests can return matching content fragments.
- Check that
collectionIdscorrectly associates the knowledge base collection related to cloud billing, avoiding retrieval scope covering irrelevant data.
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.