What the Data for This Category Looks Like
Internal office data for refined oil sales comes primarily from four sources: daily operation ledgers of gas stations (including fields such as oil tank number, fuel grade, inventory in liters, daily sales volume), employee operation manuals (including fueling procedures, equipment maintenance steps, safety specifications), business system documents (including attendance rules, expense reimbursement procedures, fuel card application rules), and equipment design drawings and maintenance logs.
Update rhythms vary widely: operation ledgers are updated daily, operation manuals are updated quarterly, business system documents are updated irregularly alongside policy adjustments, and equipment logs are generated in real time during maintenance operations. Document structures include two categories: structured tables and unstructured text. Fields involve specialized units such as liters, tons, yuan per liter, and shifts.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
The multi-type data characteristics of refined oil sales impose clear constraints on retrieval and recall: Structured operation ledgers contain precise fields and specialized units. Retrieval must prioritize matching field semantics to avoid incorrect results from unit ambiguity. Long unstructured operation manuals and system documents require reasonable control of chunk length to retain complete context. Differences in data update frequencies require limiting retrieval scope to the corresponding dataset to avoid mixing unsynchronized old data. Highly specialized equipment drawing documents need recall optimization using professional terminology to ensure matching meets internal service requirements. Additionally, internal employee questions are often colloquial. Balancing exact matching and semantic expansion is necessary to avoid missing key business information.
How to Configure the Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Aligns with semantic matching accuracy for specialized refined oil sales documents, filters low-match irrelevant content, and avoids returning incorrect fuel parameters or operation steps |
limit | Top 8 entries | Controls the number of returned reference documents, prevents internal employees from receiving excessive redundant information, and covers common office and service scenarios |
searchMode | hybrid | Combines structured field matching and semantic retrieval, adapts to mixed data scenarios that include both operation ledgers and operation manuals |
usingReRank | Enabled | Re-ranks initially recalled results to prioritize documents more relevant to employee questions, improving retrieval accuracy |
datasetSearchUsingExtensionQuery | Enabled | Optimizes colloquial employee questions by expanding them, for example, converting "How much fuel is left in the tank today" into a precise query matching inventory ledger fields |
collectionIds | Fill in the collection IDs of bound gas station operations, operation manuals, and other relevant collections based on actual conditions | Precise retrieval scope limitation, avoids mixing irrelevant public knowledge base content, and ensures recalled results only target internal office scenarios |
Three Common Configuration Mistakes
- Retrieval time increases significantly. The symptom is that knowledge base retrieval logs show elapsed time exceeding 10 seconds, and the number of returned results far exceeds the set value. The cause is that
collectionIdsis not configured, causing the system to scan the full knowledge base collection and load irrelevant non-operation documents additionally. - Front-end cannot display retrieved document names. The symptom is that the document list on the front-end page only shows content snippets without corresponding document titles. The cause is that
searchModeis not set to hybrid, so structured ledger field information is not correctly extracted and displayed. - Reference variables display as empty. The symptom is that after selecting a target knowledge base, there are no optional options in the reference variable dropdown menu. The cause is that
datasetSearchUsingExtensionQueryis not enabled, so the system cannot recognize the knowledge base's field structure and cannot generate usable reference variables.
How to Verify Successful Configuration
- Execute test queries such as "Current inventory of No. 92 fuel tank", check if returned documents include precise fields such as oil tank number, fuel grade, inventory in liters, and if units match correctly.
- View retrieval time logs, confirm that the number of returned documents does not exceed the value set by the
limitconfiguration, and verify that the retrieval scope complies with the limitations ofcollectionIds. - Test colloquial queries such as "How to apply for an employee fuel card", check if corresponding operation procedure documents are returned, and confirm that
datasetSearchUsingExtensionQueryis effective. - Click the document name displayed on the front-end, confirm that it can jump to the complete document page, and that the document content matches the query relevance as expected.
The values provided here are common starting points and should be validated against relevant sample datasets.
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.