What data for this category looks like
Civil aviation airport policy and regulatory basis data comes from regulations and normative documents issued by the Civil Aviation Administration of China, local regulatory requirements from regional civil aviation administrations, and the airport's own operation support systems and service specification documents. Data updates happen in real time when higher-level documents are released or internal regulations are revised, with no fixed cycle. Most documents use a chapter-based structure, including clause numbers, applicable scenarios, and execution standards. Metadata fields include document number, issuing unit, effective date, applicable airport category, and units such as clauses, hours, and passenger trips.
What constraints do these characteristics impose on the "knowledge base retrieval and recall" link?
The chapter-based structure of policy documents requires retaining clause hierarchy during retrieval. Splitting documents without this hierarchy will lose contextual association, leading to retrieved results being taken out of context. The presence of metadata fields requires configuring filtering rules in the retrieval link. These rules exclude expired or inapplicable policy content. The high-frequency, non-fixed update rhythm requires the knowledge base synchronization process to adapt to real-time or near-real-time update requirements. The existence of multi-source policies — general civil aviation rules separated from local regulatory requirements — creates a need for parallel retrieval across multiple knowledge bases. Clear retrieval priorities for different knowledge bases are required.
How to configure the settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Civil aviation policy clauses are rigorous. This range balances recall coverage and precision, filtering low-relevance non-policy content |
limit | Top 10 entries | Single-round retrieval results should not be excessive, to avoid context overflow and meet the concise needs of policy Q&A |
searchMode | Full-text retrieval | Most policy documents use a chapter-based structure. Full-text retrieval can cover associated expressions within clauses, improving matching completeness |
usingReRank | Enabled | Rerank retrieved results to correct local deviations in semantic similarity, improving matching accuracy for policy clauses |
collectionIds | Two knowledge base IDs configured per local classification | Distinguish between general civil aviation policy and local regulatory knowledge bases, meeting the priority needs of multi-source retrieval |
datasetSearchUsingExtensionQuery | Enabled | Perform semantic expansion on user questions to cover synonymous expressions in policy clauses, such as matching "flight delay support" to relevant policy clauses |
Three common misconfigurations
- Phenomenon: Knowledge base retrieval returns "No matching content", but the corresponding policy exists in the knowledge base. Cause: Multi-knowledge base fallback retrieval logic is not configured correctly. Only the first configured knowledge base is called, and subsequent retrieval is not triggered.
- Phenomenon: Expired policy documents appear in retrieval results. Cause: Metadata filtering rules based on effective date are not configured, resulting in retrieval of expired policy clauses.
- Phenomenon: The number of retrieved results does not match the configured
limit. Cause: The retrieval return upper limit parameter is not set correctly, or the specified number of results is not retained after reranking operations.
How to confirm the configuration is correct
- Upload a test policy file marked with an expired effective date, construct a corresponding question, and verify that the retrieved results do not include the content of this file.
- Construct test questions corresponding to general civil aviation policies and local regulatory requirements respectively, and verify that the results come from the corresponding configured knowledge bases.
- Adjust the
similarityparameter value, observe changes in the relevance of retrieval results, and confirm that the parameter configuration takes effect. - Construct a question that only exists in the second knowledge base, and verify that when the first knowledge base has no matching results, retrieval of the second knowledge base is automatically triggered.
The values provided 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.