What the Data for This Category Looks Like
Urban gas ledger data mainly comes from three types of systems: station inspection logs, user payment archives, and pipeline network operation and maintenance records. Data update rhythms follow these rules: inspection data is synced to the platform in real time, payment ledgers are updated daily, and pipeline network basic data is calibrated monthly. Single ledger documents are mostly Excel files with multiple worksheets or docx files with multi-level directories. They include fields such as station number, inspection time, pipeline pressure value, user account address, and payment amount. Units include kilopascals, cubic meters, CNY, and others.
Constraints Imposed on Knowledge Base Retrieval and Recall
Multi-worksheet Excel and multi-level directory docx ledgers mean basic retrieval cannot cover all content in subsheets or subdirectories. Retrieval mode must be adjusted to support cross-structure recall. Fields contain a large number of professional units and identifying numbers. Retrieval must match both field names and units to return valid results. Otherwise, irrelevant general gas knowledge may be returned. High-frequency real-time inspection data and monthly-calibrated pipeline network data require retrieval to adapt to data sources with different update frequencies. This avoids recalling expired or unsynced content. Single documents have wide length ranges. The chunking and recall logic must adapt to both short-text inspection records and long-text reports.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.65–0.75 | Adapts to the matching accuracy of professional fields in gas ledgers, avoids recalling irrelevant general gas documents |
limit | Top 8–12 entries | Covers multiple paragraph chunks from single long documents, while keeping context window usage under 12000 tokens |
searchMode | Full-text retrieval | Supports recall of content across worksheets and multi-level directories, adapts to multi-format ledger structures |
usingReRank | Enabled | Re-ranks recalled results by relevance, filters out results that match fields but are semantically irrelevant |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval keywords to match identifying fields such as station numbers and user numbers in ledgers |
collectionIds | Fill in according to the dedicated gas ledger dataset | Isolates knowledge base data from other categories, avoids recalling irrelevant non-gas content |
Three Common Configuration Mistakes
- Phenomenon: Calling the knowledge base retrieval interface returns an unsupported format error. Browsing the associated link normally downloads the docx file. Cause: No extended retrieval adaptation is configured for multi-worksheet Excel or multi-level directory docx files, or the interface does not correctly handle embedded document content.
- Phenomenon: Retrieval results contain a large amount of content unrelated to gas ledgers, and the number of recalled entries far exceeds the set limit. Cause:
collectionIdsis not specified as the dedicated gas ledger dataset, or thesimilaritythreshold is set too low. - Phenomenon: Subdocuments under multi-level directories are not retrieved, only root directory documents are recalled. Cause:
searchModeis not set to full-text retrieval mode, ordatasetSearchUsingExtensionQueryis not configured to support multi-level directory traversal.
How to Confirm Configuration Is Correct
- Upload a multi-level directory docx gas ledger document, perform a retrieval test, and confirm that content in subdirectories is successfully recalled.
- Submit a station number as a retrieval keyword, confirm that returned results include matching professional fields such as station numbers and pressure values, and irrelevant results are effectively filtered.
- View the retrieval log, confirm that configuration parameters such as
searchModeandusingReRankare correctly loaded, and no parameter conflict prompts are present. - Simulate an operation to skip retrieval, confirm that the retrieval link is not triggered, and the process directly proceeds to the subsequent ledger summary and report generation stage.
Note: The values listed here are common starting points. Adjustments should be made based on relevant 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.