What This Category of Data Looks Like
This category of data typically comes from the 12345 citizen hotline platform, contracting unit business systems, official knowledge bases, follow-up evaluation data, and local hotline management regulations. Daily processed documents include request registration forms, dispatch records, handling responses, follow-up records, extension applications, return explanations, and knowledge base entries. Data is updated daily. Core fields include request number, request type, dispatching unit, response content, follow-up results, and other related fields. There are no fixed numerical indicators with a fixed statistical caliber.
Constraints on Knowledge Base Retrieval and Recall
Multiple data sources create differences in field formats across different documents. During retrieval, match associated fields such as request number and dispatching unit to avoid retrieving irrelevant tickets. The daily update rhythm requires the retrieval link to sync the latest dispatch and response data. It cannot rely on expired caches. Diverse document types cover the full process from request registration to follow-up, with a wide span of semantic scenarios. Precise retrieval of historical handling reference content matching the request type is required.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.65–0.75 | Matches semantic similarity between request types and response content, covering retrieval ranges for common request scenarios |
limit | Top 8 entries | Controls the number of knowledge base entries retrieved per call, avoiding context overload that reduces ledger summary efficiency |
searchMode | Semantic Retrieval | Adapts to semantic matching needs for multi-source ticket documents, making up for insufficient accuracy of keyword retrieval |
usingReRank | Enabled | Performs secondary sorting on retrieval results to filter low-relevance historical response content, improving the reference value of ledger summaries |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands retrieval keywords to cover synonymous expressions of request types, adapting to diverse public request phrasing |
collectionIds | Bind to the 12345 citizen hotline exclusive knowledge base collection | Limits retrieval scope to only official documents related to ticket processing, avoiding retrieval of irrelevant content |
Three Common Configuration Mistakes
- After calling the knowledge base retrieval API, the associated knowledge base file list cannot be obtained. The cause is that
collectionIdsis not properly configured to limit the retrieval scope, resulting in retrieval results that are not associated with file information from the corresponding knowledge base collection. - The number of results returned by retrieval is far lower than expected. The cause is that the
similaritythreshold is set too high, filtering out historical ticket content that meets business requirements. - A timeout error occurs during the ledger summary process. The cause is that
usingReRankis not enabled andlimitis set too large, resulting in excessive redundant content participating in subsequent text processing links.
How to Verify Correct Configuration
- View the knowledge base retrieval configuration page, confirm that
collectionIdsis bound to the 12345 citizen hotline exclusive document collection - Enter a typical public request content for testing, check whether the number of retrieval results matches the
limitsetting - Check the sorting logic of retrieval results, confirm that
usingReRankis active, and results are sorted from highest to lowest relevance - View debug logs or interface return content, confirm that retrieval automatically expands query keywords to cover synonymous expressions of the request
The values provided here 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.