What This Category of Data Looks Like
Data sources include local gas group official service guides, pipeline operation and maintenance archives, housing and urban-rural development department gas safety regulations, and user frequently asked questions. Updates follow a quarterly regular schedule, with real-time updates when policies are adjusted or pipeline reconstruction occurs. Documents are structured by business modules, including service items such as account opening, repair requests, and payment, parameters such as pipeline pressure and pipe diameter, and safety operation regulations. Fields include item code, processing time limit, required materials, and charging standards. Units include cubic meters, yuan per cubic meter, working days, and others.
Constraints for Knowledge Base Retrieval and Recall
Urban gas data contains precise professional terminology and structured parameters. Retrieval must match terminology accurately. Semantic drift will lead to incorrect service guidance otherwise. Most users submitting service inquiries use colloquial language. Teams must match these colloquial questions to corresponding structured documents. Parameter units and fields must not be confused. Datasets are divided by business modules. Retrieval scope must be limited to avoid cross-category content. All recalled results must relate only to gas services. Data has long update cycles but rigorous content. The latest document versions must be prioritized for recall. Expired content must not be used.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Colloquial questions are common for urban gas service inquiries. A threshold that is too low will include irrelevant content, while a threshold that is too high will miss accurately matched service procedure documents |
limit | Top 8 entries | Gas-related documents cover multiple categories including service procedures, safety regulations, and pipeline parameters. 8 entries can cover all necessary information required for user questions and avoid overly long context |
searchMode | Semantic + keyword hybrid retrieval | Colloquial questions require semantic matching, while precise fields such as pipeline parameters and charging standards require keyword matching. Hybrid mode balances recall scope and accuracy |
usingReRank | Enabled | Gas content contains a large number of professional terms. Reranking filters out low-quality results that are semantically related but do not match the terms, improving the availability of final recalled content |
datasetSearchUsingExtensionQuery | Enabled | User questions often omit professional prefixes such as omitting "household" before "gas meter". Extended queries can supplement missing qualifying terms to improve matching accuracy |
collectionIds | Bind three dedicated datasets: gas service guides, pipeline operation and maintenance archives, and safety regulations | Dividing datasets by business modules avoids mixing cross-category content and improves retrieval targeting |
Three Common Misconfigurations
- Symptom: A 503 error is prompted when uploading gas service documents via the dialog box, while the knowledge base file upload process works normally. Cause: The dedicated
collectionIdsare not specified. The system attempts to assign documents to the general dataset, triggering the concurrent upload threshold limit of the general dataset. - Symptom: When semantically retrieving "household gas meter replacement procedure", documents containing the keyword "replacement procedure" in the payment section are not recalled. Cause:
searchModehybrid retrieval is not enabled, and only pure semantic retrieval is used. Keyword fields are not matched, resulting in documents with clear keywords not being recalled. - Symptom: When using the association analysis function, cross-document pipeline connection relationships cannot be recalled. Cause:
datasetSearchUsingExtensionQueryis not enabled, and professional term related words are not supplemented, resulting in failed relationship extraction.
How to Verify Correct Configuration
- Submit a colloquial gas service inquiry such as "How do I pay my household gas bill?" and check whether the recalled results include the corresponding payment procedure document.
- View the retrieval logs in FastGPT v4.17.0 to confirm that
searchModeis set to hybrid mode,usingReRankis active, and the number of recalled entries matches the presetlimitvalue. - Upload a test document to the dedicated dataset, confirm that
collectionIdsare bound correctly, and that the document can be retrieved normally. - Adjust the
similaritythreshold, test recall results under different thresholds, and confirm that matching accuracy meets business requirements.
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.