What This Category's Data Looks Like
Data sources for airport service inquiries mainly include airport official operation manuals, real-time flight scheduling systems, terminal service announcements, security check and baggage handling specification documents, and special service application guidelines. The data update schedule falls into two categories: fixed-rule documents are updated weekly or monthly, while real-time flight and temporary control notifications are synced hourly or in real time. Most single documents combine structured fields, including service category, applicable terminal, effective time period, inquiry contact information, and more. Units include minutes, kilograms, terminal numbers, flight numbers, and others.
What Constraints These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
First, real-time data accounts for a large share. The retrieval link must support low-latency recall to avoid poor inquiry experiences caused by long wait times. Second, documents contain multi-dimensional structured fields. Targeted matching of corresponding retrieval dimensions is required to avoid returning invalid content across terminals or beyond effective time periods. Third, high-frequency precise fields in inquiries such as contact information and flight numbers may have matching deviations if relying solely on semantic retrieval. Combine full-text retrieval to enhance precise recall. Fourth, temporary announcement documents need to have effective time periods marked during indexing to filter expired content and avoid misleading inquirers. Fifth, single documents vary greatly in length. Long documents require proper chunking to prevent key information from being truncated during retrieval.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the field precision of airport documents, filters low-relevance cross-terminal and outdated content, and balances recall relevance and recall volume |
limit | Top 8 entries | Balances the information volume of recall results and large model context window limits, reduces single retrieval computation, optimizes retrieval time, and adapts to information combination needs of airport inquiry scenarios |
searchMode | hybrid | Combines semantic and full-text retrieval, covers recall needs for precise fields such as flight numbers and contact information, and solves the problem that pure semantic retrieval cannot match structured text |
usingReRank | Enabled | Re-ranks recall results by relevance, resolves the problem of mixed multiple results, and improves the efficiency of information acquisition for inquirers |
datasetSearchUsingExtensionQuery | Enabled | Optimizes user inquiries by adding associated queries for airport-specific terms, such as adding "boarding gate processing" to "check-in counter", to improve recall coverage |
collectionIds | Specify airport-specific knowledge base collections | Isolates knowledge base data from different airports, avoids interference from cross-regional content, and ensures recall results only relate to the currently served airport |
Three Common Misconfigurations
- Phenomenon: Precise fields such as imported airport service phone numbers and flight contact numbers cannot be recalled by both semantic retrieval and full-text retrieval, and irrelevant terminal guidance content is returned. Reason: Hybrid retrieval mode is not enabled, and relying solely on semantic retrieval cannot match structured contact information fields. Full-text retrieval needs to be combined to cover precise text matching.
- Phenomenon:
usingReRankis configured and index re-ranking is enabled, but no change in sorted results is seen during online recall testing. Reason: The re-ranking switch is not enabled synchronously in the retrieval link. Only configuring it during the indexing phase does not take effect. Confirm that the re-ranking function is enabled in the retrieval parameters. - Phenomenon: After setting
limitto 3000, retrieval results cannot be passed into the large model context window. Reason: The total length of single recalled content exceeds the context upper limit supported by the large model. Adjust the number of recalled entries or the length of single document chunks to avoid context overflow.
How to Confirm Proper Configuration
- Upload a test document containing an airport service phone number, initiate an inquiry that includes the number, and confirm that the document is included in the retrieval results.
- After enabling
usingReRank, compare the initial sorting of recall results with the re-sorted results, and confirm that the order of the results has changed. - Adjust the
limitparameter to different values, and check whether the number of returned retrieval results matches the set value. - View the associated configuration of the knowledge base collection, confirm that
collectionIdsis bound to the current airport's exclusive knowledge base, and no documents from other regions are mixed in.
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.