What the Data for This Category Looks Like
Data is primarily sourced from gas station operation manuals, fuel retail price lists, fuel card business handling procedures, customer service standard documents, daily updated oil price adjustment notices, and weekly updated station promotional activity and shift schedule information. Most documents use a structured, chaptered format, with clear fields including fuel grade, price unit yuan per liter, business handling steps, and consultation response specifications. Update rhythms fall into two categories: real-time synchronization for daily price changes, and regular revisions for business operation processes.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
Structured fuel price lists and real-time oil price data require retrieval to accurately match fields such as unit price and grade, to prevent information deviation from fuzzy recall. Chaptered operation documents need targeted recall by business module, to avoid mixing irrelevant cross-business content in responses. Real-time updated oil price changes require the knowledge base sync frequency to align with daily price adjustment cycles; otherwise, recalled price information will become invalid. The multi-field document structure also requires support for filtering recall results by station type and fuel category, to ensure responses fit the consultation scenario.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches structured fuel price lists and business procedures, avoids introducing irrelevant results with too low a value, and avoids missing valid business explanations with too high a value |
limit | Top 8 entries | Covers common business types for gas station consultations, while controlling context length to fit large model processing capabilities |
searchMode | keyword+vector | Balances precise price list retrieval via keyword matching and business scenario understanding via vector recall, adapting to multiple types of consultation needs |
usingReRank | Enabled | Rearranges multiple recalled results by relevance, ensuring core information such as oil prices and business steps appears at the top of responses |
datasetSearchUsingExtensionQuery | Enabled | Automatically optimizes colloquial consultations into structured queries, matching exclusive fields of price lists and business procedures |
collectionIds | Bound to the exclusive collection of gas station business knowledge base | Isolates knowledge bases of different stations or business modules, avoiding cross-category data interfering with retrieval results |
Three Common Misconfigurations
- Phenomenon: When batch uploading multiple docx and pdf format business documents, some files fail to complete indexing. Cause: Parallel batch creation is not used, single-threaded processing speed is insufficient, and cannot cover multi-file simultaneous upload tasks.
- Phenomenon: When retrieving colloquial user consultations such as "How much is No. 92 gasoline today", the returned results have low matching degree with actual needs. Cause:
datasetSearchUsingExtensionQueryis not enabled, colloquial questions are not optimized into structured queries, and cannot accurately match exclusive field information of price lists. - Phenomenon: Retrieval requests return timeout errors, and the interface shows request failure. Cause: The
limitparameter is set too high, the number of recalled documents is excessive, exceeding the context processing limit of the large model, leading to processing timeout.
How to Verify Successful Configuration
- Upload a test fuel price list document, initiate the consultation "What is the current price of No. 95 gasoline", and check whether the returned result contains the accurate unit price value.
- Initiate a batch upload test, confirm that multiple docx and pdf format business documents can complete indexing simultaneously without omission.
- After binding the exclusive knowledge base via
collectionIds, initiate a cross-business consultation, and check that the returned results do not contain irrelevant administrative documents. - Adjust the value range of the
similarityparameter, verify whether the matching degree of recall results meets business requirements in different scenarios.
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.