What this category of data looks like
Refined oil sales policy and regulatory basis data mainly comes from normative documents issued by national industry authorities, local commerce and market supervision departments, and compliance execution manuals from petroleum and petrochemical sales enterprises. The document structure includes fields such as issuing entity, document number, effective date, applicable region, core clauses, and penalty clauses for violations. Some quantitative compliance requirements use standard units including liters, tons, and square meters. Update rhythm follows regulatory policy adjustments. Core policy documents have a low update frequency, while supplementary implementation rules have a high update frequency, with no fixed cycle.
What constraints these characteristics impose on knowledge base retrieval and recall
Policy documents have version differences. Invalid outdated documents must be filtered during the recall stage. Retrieval must link the effective date field for validity verification. Documents contain clear applicable region and business entity clauses. Recall results must match the region and business scenario of user questions to avoid cross-regional compliance risks. Core clauses are mostly formal compliance requirements. Retrieval results must accurately correspond to the original text of clauses to avoid compliance issues caused by semantic drift. Some supplementary implementation rules have a high update frequency. The knowledge base must be synchronized regularly to prevent recall of outdated execution standards.
How to set configurations
These configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Refined oil policy clauses use precise wording. This range balances recall completeness and accuracy, filtering irrelevant results with low matching degrees |
limit | Top 10 entries | The number of policy clauses is limited. Too many results increase context processing burden. 10 entries can cover core relevant clauses |
searchMode | Exact match | Policy retrieval requires accurate correspondence to the original text of clauses. Exact match mode avoids irrelevant semantic interference and improves compliance retrieval accuracy |
usingReRank | Enabled | Reranking recall results prioritizes content with higher matching degrees to policy clauses, optimizing retrieval result sorting logic |
datasetSearchUsingExtensionQuery | Disabled | Policy queries do not require extended searches. Direct matching of original text avoids introducing irrelevant content and ensures retrieval result rigor |
collectionIds | Specify the exclusive collection ID for compliance policies | Avoid mixing non-policy knowledge base content, ensuring the retrieval scope only covers refined oil sales-related policy documents |
Three common errors
- Symptom: A prompt stating
invalid configuration parameter name "hnsw.max_scan_tuples"appears during retrieval, or no results are returned. Cause: Custom vector retrieval parameters not supported by FastGPT are used, and the specified configuration item list is not adopted. - Symptom: Recall results include outdated expired policy documents. Cause: The effective date field is not linked in the retrieval configuration for validity filtering, leading to expired clauses being included in the recall scope.
- Symptom: Retrieval results mix policy content unrelated to refined oil sales. Cause: The
collectionIdsparameter is not correctly specified. The retrieval scope covers irrelevant knowledge base collections, leading to recall of cross-category content.
How to confirm configuration is complete
- Submit a query targeting specific refined oil retail compliance clauses, and check if returned results include the original text of the corresponding policy.
- Review the parameter values of the retrieval configuration, and confirm that
similarity,limit, and other configuration items meet preset rules. - Verify that the retrieval scope only covers the exclusive knowledge base collection for refined oil sales policies, avoiding mixing irrelevant content.
- Submit a keyword test for an expired policy, and confirm that the result is filtered with no expired clauses returned.
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 2026-09-28.