What This Category's Data Looks Like
The data for urban gas application material and case review primarily comes from internal business systems of gas operating enterprises. It includes user installation application forms, site inspection records, completion acceptance cases, payment vouchers, and similar documents. Data updates in real time alongside business work orders. New applications and newly archived cases trigger data creation and synchronization. Each individual document uses a fixed structure, with fields including the application subject’s unified social credit code, gas usage address, designed gas usage scale, construction plan, and approval circulation records. Field units include cubic meters per hour, square meters, yuan, and others.
What Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
First, data updates in real time alongside business work orders. This requires the retrieval pipeline to support incremental recall, to avoid retrieving expired and invalidated cases. Second, documents contain specialized fields and units. Retrieval must match field semantics and unit consistency, otherwise irrelevant results may appear. Third, individual documents have a fixed structure and clear field boundaries. This allows targeted configuration of chunking rules to avoid semantic confusion across fields. Fourth, approval circulation records across different cases have nested levels. Recall must retain hierarchical associations to ensure the logical integrity of retrieval results.
Configuration Settings
The following configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Urban gas cases contain a large number of specialized terms. This similarity range filters low-relevance results while balancing recall rate and precision |
limit | Top 8–12 entries | A single case has numerous approval circulation records. A small number of entries avoids context overload while covering core approval information |
searchMode | Semantic Retrieval | Documents have clear field boundaries. Semantic retrieval accurately matches specialized content, avoiding ambiguity from keyword matching |
usingReRank | Enabled | There are many specialized terms. Reranking corrects semantic bias in initial retrieval and improves result relevance |
datasetSearchUsingExtensionQuery | Enabled | Queries for application materials often contain vague expressions. Extended queries cover synonymous terms and expand recall coverage |
collectionIds | Bind dedicated knowledge base collections for gas installation and cases | Isolate document data from other categories to avoid irrelevant content in retrieval results |
Three Common Configuration Mistakes
- Issue: Uploaded gas application material files cannot be read during conversations, but file uploads function normally in the knowledge base backend. Cause: When deployed locally, the file storage volume was not mounted correctly, or incorrect file read permissions were configured. This prevents the conversation context from accessing temporarily uploaded files.
- Issue: Knowledge base retrieval occasionally returns empty results, but relevant content can be successfully recalled during knowledge base backend testing. Cause: The
similaritythreshold is set too high, andusingReRankis not enabled. Initial retrieval results are filtered out, leaving no valid content. - Issue: Queries unrelated to gas application content return case approval records. Cause: Dedicated knowledge base collections were not bound via
collectionIds, so the retrieval scope includes document data from other categories.
How to Verify Proper Configuration
- Submit a query targeting a specific gas usage address. Check that returned results include the application material fields matching that address, to verify the accuracy of the retrieval scope.
- Upload a new completion acceptance case. Trigger knowledge base incremental synchronization, then submit a corresponding query to confirm the new data is successfully recalled.
- Adjust the
similarityparameter range. Verify that the relevance of returned results aligns with business expectations. - Disable
datasetSearchUsingExtensionQuery, compare the coverage of retrieval results across two attempts, to confirm the extended query configuration is active.
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.