What data for this category looks like
Data for this category originates primarily from cloud management platforms, order systems, IAM permission systems, and billing systems. Updates are synced in real time per order. Daily documents include resource activation forms, instance configuration tables, account permission forms, and delivery confirmation forms. Fields cover instance specifications (CPU core count, memory capacity, GPU model), permission scope, billing items, order numbers, delivery status, and similar details. Common units include cores, GB, units, and yuan. Document structures are split by business processes. Individual data entries are small in size, but have strong field correlation.
What constraints do these characteristics impose on knowledge base retrieval and recall
Cross-source data synchronization requirements from multiple system sources mandate that the knowledge base be configured with cross-dataset associated recall rules. Adapting to real-time update features requires enabling incremental synchronization mechanisms. The document structure split by business processes requires that retrieval keywords match business scenario tags to avoid invalid recall of cross-category documents. The strong field correlation characteristic requires limiting field matching ranges during retrieval to prevent mismatches of irrelevant fields such as specifications and permissions. Additionally, errors directly impact customer business deployment and billing accuracy, so the precision of recall results must be improved to reduce the return of irrelevant content.
How to set configurations
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
similarity | 0.75–0.85 | This category has strong field correlation, so low-similarity irrelevant documents must be filtered to avoid returning non-cloud host provisioning content |
limit | Top 8 entries | Daily document volume is small, 8 entries cover all relevant business documents and avoid context overload |
usingReRank | Enabled | Multi-field associated documents require re-ranking to raise the display priority of highly relevant results |
datasetSearchUsingExtensionQuery | Enabled | Can automatically supplement business keywords such as "cloud host" and "resource activation" based on user questions to improve retrieval matching accuracy |
collectionIds | Bind knowledge base collection IDs related to cloud host provisioning | Limit retrieval scope to exclude interference from documents of other categories |
searchMode | Vector retrieval + keyword hybrid | Balance semantic matching and precise keyword matching to adapt to business documents with strong field correlation |
Three common mistakes
- Symptom: Knowledge base documents are retrieved, but the large model response is overly concise and does not return specific clause content. Cause:
datasetSearchUsingExtensionQueryis not enabled, and the full original document text is not included in the context, only abstract information is extracted. - Symptom: Knowledge base search takes too long, and single queries time out. Cause: The
collectionIdsretrieval scope is not limited, the retrieval scope covers the entire platform knowledge base, and thelimitvalue is not adjusted, resulting in excessive redundant documents to process. - Symptom: Recall results include non-cloud host provisioning documents, such as cloud network configuration content. Cause:
collectionIdsis not configured correctly, and irrelevant knowledge base collections are bound, resulting in recall of cross-category documents.
How to confirm configurations are correctly set
- View the retrieval logs in FastGPT v4.17.0, check whether the document sources of the recall results are all collections related to cloud host provisioning, to confirm that the
collectionIdsconfiguration takes effect. - Initiate a query containing specific specification parameters, check whether the returned results include the complete content of the corresponding instance configuration tables and resource activation forms, to confirm the configuration effects of
datasetSearchUsingExtensionQueryandusingReRank. - Simulate routine business queries, verify whether retrieval latency meets business expectations, and adjust the values of
similarityandlimitto balance retrieval efficiency and precision. - Compare retrieval results of the same query across multiple attempts, confirm the consistency of recall results, and verify configuration stability.
The values provided here 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.