What the data for this category looks like
The data sources for urban gas business applications mainly include official business manuals, online application guides, offline business hall public documents, and latest policy documents. The document update schedule follows business adjustments or policy changes. Regular business processes are updated every 6 months, and emergency policy changes can trigger updates at any time. Single documents have a fixed structure, including fields such as application conditions, required materials, step-by-step process, charging standards, processing time limits, and consultation channels. Charging standards mostly use units such as "yuan/cubic meter" or "yuan/household". Processing time limits use "working days" as the unit. Material lists clearly note quantity requirements.
What constraints these characteristics impose on knowledge base retrieval and recall
Fixed document structure requires retrieval results to be sorted by business logic to avoid chaotic steps. Varied document formats from multiple sources require unified segmentation rules to ensure retrieval consistency. Clear field and unit requirements mean unit information must be matched during retrieval, preventing users from obtaining incorrect charging or time limit data. Embedded flow diagrams in documents require the knowledge base to support associated recall and display of images. Infrequent but urgent update needs require a flexible synchronization mechanism to ensure data timeliness.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Urban gas business processes have high requirements for terminology accuracy. This threshold filters irrelevant results while retaining valid recalled content |
limit | Top 8 entries | The number of steps in a single business process document usually does not exceed 10. Recalling the top 8 entries covers the complete process without exceeding the context window |
searchMode | strict | Avoids mismatches between similar terms such as "account opening" and "card opening", and ensures retrieval results are strictly aligned with business terminology |
usingReRank | Enabled | Reordering arranges business process steps in logical order, improving the fluency of information acquisition |
datasetSearchUsingExtensionQuery | Enabled | Supports expanding colloquial user questions (such as "how to open a gas account") into standard business terminology, improving recall coverage |
collectionIds | Bind to the dedicated gas business application knowledge base | Isolates knowledge base content from other categories to avoid irrelevant information mixing into retrieval results |
Three Common Mistakes
- Phenomenon: After uploading javadoc-formatted HTML interface documents, the knowledge base fails to parse any content. Cause: Structured parsing rules for HTML documents are not enabled. The HTML file is only uploaded as plain text, and interface information and associated resources are not extracted.
- Phenomenon: Embedded gas flow diagrams in search results fail to load normally. Cause: Only the HTML file was imported when uploading the document, but associated image resources were not uploaded synchronously, or the image links were not correctly bound to the corresponding segmented text blocks.
- Phenomenon: Existing knowledge base content is lost after upgrading FastGPT. Cause: Data was not migrated via the knowledge base export and import function, and the original storage directory was overwritten by directly pulling the image.
How to Confirm Configuration Is Valid
- Upload a standard gas account opening process test document, search for "what materials are required for gas account opening", and verify that the recalled results include clear fields such as required materials and processing time limits.
- View the segmented text blocks in the knowledge base, confirm that image links are correctly embedded, and see associated display items for images when searching test terms.
- Adjust the
similarityparameter to 0.6 and 0.9, test the change in the number of retrieval results separately, and confirm that the parameter configuration takes effect. - Export the current gas business application knowledge base, upgrade to FastGPT v4.17.0, then re-import it, and confirm that the knowledge base content is complete and not missing.
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.