What this category of data looks like
This category of data typically originates from the 12345 citizen hotline platform, contracting entity business systems, official knowledge bases, and return visit evaluation data. The update frequency is daily. Daily documents include request registration forms, dispatch records, handling responses, delay applications, return explanations, and policy knowledge base entries. Fields include request number, contracting entity, request type, response content, and return visit result. Each data entry forms a complete closed-loop link around citizen requests.
What constraints these characteristics impose on knowledge base retrieval and recall
Daily updated data requires the retrieval link to support incremental synchronization, preventing excessive resource use from full indexing. Multiple ticket document types have differences between colloquial requests and standardized policy text, so the system must adapt retrieval logic for mixed text. Data forms a closed loop around citizen requests, so retrieved content must link to the dispatch and response history of the corresponding request. The retrieval scope must also be limited to 12345-specific policies and business documents to avoid interference from irrelevant content. Errors in ticket handling relate to issues such as overdue deadlines and reprocessing, so retrieved results need high matching accuracy to prevent ineffective retrieval from delaying handling.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the semantic difference between colloquial 12345 requests and standardized policy text, balances retrieval accuracy and coverage |
limit | Top 8 entries | Adapts to the single-response needs of ticket handling, avoids excessive context exceeding processing limits |
searchMode | hybrid | Combines keyword and semantic retrieval, covers keyword matching in requests and precise matching of policy basis |
usingReRank | Enabled | Reorders retrieved results, filters low-matching policy entries, improves response accuracy |
datasetSearchUsingExtensionQuery | Enabled | Generates extended queries based on original requests, covers implicit policy needs in requests |
collectionIds | Exclusive 12345 ticket and policy dataset IDs | Limits retrieval scope to exclusive datasets, avoids interference from irrelevant content |
Three common mistakes
- Phenomenon: No response when uploading multiple files, and a network failure error appears after long transcoding for a single file upload. Cause: The file upload threshold for the corresponding dataset is not configured, or the storage quota of the local deployment environment is insufficient, causing timeout for large or multiple file uploads.
- Phenomenon: Image understanding or indexing model options do not appear when creating a new knowledge base. Cause: FastGPT versions below v4.17.0 do not include this feature built-in, and local deployments lack the additional installation of the corresponding model package. Only commercial deployments include this configuration by default.
- Phenomenon: Retrieved recall results do not include knowledge base citation sources in the response body. Cause: The configuration switch for knowledge base citation output is not enabled, or fixed format requirements for citation sources are not included in the prompt.
How to confirm the configuration is correct
- Upload one request text containing policy keywords, check whether the number of retrieved results matches the value set in the
limitconfiguration. - Check the dataset retrieval scope, confirm that only 12345-exclusive ticket and policy documents are included, with no external irrelevant datasets.
- Initiate a test query, check whether the returned results have been reordered and linked to the corresponding knowledge base sources.
- Simulate batch document upload, confirm that the upload process has no timeout or network failure errors, and meets the configured upload threshold requirements.
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.