What this category’s data includes
Urban water utility policy and regulatory basis data originates primarily from normative documents and industry standards issued by water utility and housing and urban-rural development authorities at all levels, as well as internal operation management systems from local urban investment water utility enterprises. Update cycles align with policy revision requirements: national-level policies receive centralized updates every six months to one year, while local policies and internal systems are released irregularly. Most documents are formal texts with official document numbers, with structures including issuing authority, effective date, applicable region, and clause number. Fields related to water usage quotas and charging standards use units such as liters per person per day and yuan per cubic meter.
How These Characteristics Impact Knowledge Base Retrieval and Recall
The official nature of urban water utility policy documents requires retrieval results to match official document numbers and clauses, to prevent generalized content from being included. The irregular update cycle means the knowledge base synchronization process must adapt to flexible update schedules. Most documents are long texts with clause numbers, so retrieval and recall must support locating content by clause. Specific unit fields require exact matching of associated units during retrieval, to avoid returning results with mismatched units. Some policies only apply to specific regions, so retrieval must support filtering recall scope by region.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Urban water utility policy texts are rigorous, so low-match irrelevant clauses must be filtered to avoid incorrect recall |
limit | Top 8–12 results | Individual policy clauses have clear content, and too many results will exceed the conversation context window |
usingReRank | Enabled | Policy texts are mostly long clauses, and reranking improves the sorting accuracy of recall results |
datasetSearchUsingExtensionQuery | Disabled | Water utility policy queries mostly require precise clause matching, so no query expansion optimization is needed |
collectionIds | Specify the collection ID of the target water utility policy knowledge base | Isolate knowledge base content across categories to avoid recalling irrelevant documents |
searchMode | hybrid | Balances keyword matching and vector matching, covering different retrieval points such as document numbers and clause content in policy texts |
Three Common Configuration Mistakes
- Issue: Recall results include non-water utility policy documents. Cause: The
collectionIdsparameter is not configured, and the target water utility policy knowledge base collection is not specified. - Issue: Recall results contain duplicate policy clause content. Cause: Duplicate uploads of the same policy document in the knowledge base, or no deduplication logic was executed during chunk processing.
- Issue: The number of search results does not match the configured
limitvalue. Cause: Thesimilaritythreshold is set incorrectly, filtering some eligible policy clauses, or there are not enough matching documents in the data source.
How to Verify Correct Configuration
- Manually enter a query related to water utility policies, and check whether the document source of the recall results is the specified water utility policy knowledge base collection.
- Adjust the
similarityparameter, observe changes in the match quality of recall results, and confirm that the parameter takes effect. - Disable
usingReRank, compare the sorting of recall results before and after, and confirm that the reranking function is enabled normally. - Check whether duplicate files in the knowledge base have been cleaned up, and confirm that duplicate content will not be recalled.
The values provided here 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.