What this type of data looks like
Four main sources generate civil aviation airport ledger data: flight operation management systems, air traffic control collaboration platforms, security check verification systems, and ground equipment operation and maintenance platforms. Data updates trigger based on operation nodes. Takeoff and landing times sync in real time. Teams summarize and archive daily operation and maintenance data each daily operation cycle. Each ledger document includes fields such as flight number, takeoff and landing time, passenger throughput, equipment number, and operation duration. Units include sorties, passenger counts, hours, and minutes. Organizations store data as structured tables or JSON format. Some historical ledgers exist as semi-structured text transcribed from scanned documents.
Constraints on knowledge base retrieval and recall
Mixed structured and semi-structured ledger documents require the retrieval link to support a combination of field-level precise matching and full-text retrieval. This prevents irrelevant fields from interfering with recall results. Real-time or high-frequency updated ledger data requires recall logic to align with data source update cycles. This avoids calls to expired historical data. Fields with multiple coexisting units, such as passenger counts and sorties, need unit verification rules added to the retrieval configuration. This prevents recall results where values do not match their units. Most ledger documents store data classified by business modules, including flights, operation and maintenance, and security checks. Dividing knowledge base collections by module narrows retrieval scope and improves accuracy.
Configuration settings
This guide adapts to the knowledge base retrieval configuration logic of FastGPT v4.17.0. The specific configuration items are listed below:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
Similarity threshold (similarity) | 0.75–0.85 | Civil aviation ledger data has high precision requirements. This interval filters low-relevance non-business data |
Single search limit (limit) | Top 8 entries | Ledger summary requires covering multi-dimensional business data. 8 entries balance recall volume and context length |
Search mode (searchMode) | hybrid | Supports combined structured field matching and full-text retrieval, adapting to mixed-structure ledger documents |
Re-ranking (usingReRank) | Enabled | Civil aviation ledger data has many fields. Re-ranking filters semantically duplicate recall results and improves report readability |
Query optimization (datasetSearchUsingExtensionQuery) | Enabled | Ledger queries often require supplementary business context. Optimization expands retrieval scope to cover associated data |
Collection ID (collectionIds) | Specify the corresponding collection ID by business module | Narrow retrieval scope and avoid cross-module irrelevant data interfering with recall accuracy |
Three common mistakes
- Symptom: An error prompt appears during testing, but the application runs normally after referencing the knowledge base. Cause: There is a mismatch between the model-side interface verification logic and the parameter verification logic of the FastGPT retrieval plugin. Structured field verification rules for ledger data are not correctly adapted.
- Symptom: The retrieval result only returns 1 document, and cannot recall other ledger data under the same module. Cause: The
limitparameter is set to 1, orsearchModeonly enables single-field matching, and the hybrid retrieval mode is not enabled. - Symptom: After importing the ledger knowledge base, the knowledge base option in the application is empty and requires manual re-selection. Cause: The
collectionIdsparameter is not specified in the configuration. The system cannot automatically associate the knowledge base collection of the corresponding business module.
How to verify a complete configuration
- Run a single ledger retrieval test, and check whether the fields and units of the recall results conform to the ledger specifications of the current business module.
- Check the number of retrieval results, and confirm whether they match the value range of the
limitparameter. - Check the
usingReRankswitch status in the retrieval configuration, and confirm whether the re-ranking function is enabled according to business requirements. - Import test expired ledger data, verify whether the retrieval result automatically filters expired content, and confirm alignment with the data source update cycle.
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 2026-09-28.