What this category of data looks like
Data sources for document and material data in the urban gas field include internal operation archives, officially released gas service specifications, pipeline operation and maintenance ledgers, and user service work order templates. Updates align with new policy implementation, pipeline commissioning, and service process adjustments.
Document structures include structured tables (such as fields on new connection application forms), long-text specifications (such as safety inspection rules), and scattered work order records.
Fields include pipeline number, gas usage address, new connection application date, charging standards, and inspection points. Units include meters, cubic meters, yuan per cubic meter, and working days.
Constraints on knowledge base retrieval and recall
There are many structured tables with fixed fields. Automatic segmentation may split a single user record into multiple text blocks. Custom separators are required to match record boundaries.
Long-text specifications have considerable length. The length of retrieved content must be controlled to fit the context window.
There are many professional synonymous expressions in the industry, such as "safety inspection" and "on-site inspection". Semantic search and re-ranking functions are needed to filter irrelevant results.
Scattered work order records have high keyword accuracy requirements. Specific fields such as pipeline numbers must be matched to avoid retrieving irrelevant content.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Urban gas professional documents have dense terminology. This threshold filters irrelevant general office documents and retains business-related content |
limit | Top 6 entries | Associated reference documents for a single gas document usually do not exceed 5 items. Too many retrievals will exceed the context window and cause confusion in large model output |
searchMode | semantic | There are many synonymous expressions in the gas industry. Semantic search can match similar inspection specifications and new connection templates |
usingReRank | Enabled | Can perform secondary sorting on retrieval results to filter documents that are semantically similar but actually irrelevant |
datasetSearchUsingExtensionQuery | Enabled | Can expand the user's vague questions to solve problems with unclear references such as "gas new connection materials" |
collectionIds | Gas business-specific knowledge base collection ID | Only retrieve documents within this collection to avoid mixing non-gas general data |
Three Common Misconfigurations
- Phenomenon: When importing an Excel-format gas user account opening template, the automatic splitting result merges entire column fields into one text block, and cannot split according to single user records. Cause: The custom separator is not set to the Excel line ending character, and the default separator cannot identify single record boundaries.
- Phenomenon: When retrieving questions related to gas inspections, the large model cannot distinguish the specific pipeline number referred to by "this pipeline", and the retrieval results do not match the query. Cause:
datasetSearchUsingExtensionQueryis not enabled, and reference resolution and question expansion are not performed. - Phenomenon: Retrieval results mix safety specification documents from the power industry, which are irrelevant to gas business. Cause:
collectionIdsis not configured as the gas-specific knowledge base collection, and cross-category documents are retrieved.
How to Verify Correct Configuration
- Upload a gas new connection Excel template, trigger the custom separator test, and check whether the split results follow the format of one entry per single user record.
- Enter vague questions such as "What is required for gas inspection", and check whether the retrieval results include relevant gas business documents.
- Confirm that the number of retrieval results matches the setting of
limit. - Compare retrieval results when
usingReRankis enabled and disabled to confirm that the re-ranking function is active.
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.