What the data for this category looks like
Usually, data sources for this category include the national integrated government service platform's "Efficiently Complete One Task" special zone, local integrated government service platforms, electronic certificate repositories, and approval systems of various departments. The update frequency is daily. Daily documents cover joint handling item lists, one-time notification sheets, joint handling application forms, material circulation sheets, parallel approval opinions, and completion receipts. Document structures include fields such as item ID, handling department, node requirements, and material list. There are certain differences in document fields across different departments, and there is no unified fixed format.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Multi-source heterogeneous cross-department data requires the retrieval pipeline to cover multiple independent datasets to avoid missing relevant joint handling item information. The daily update frequency requires configuring an incremental synchronization mechanism to ensure that retrieval results use the latest joint handling item data. Diverse document structures and differentiated fields require adapting chunking logic for different documents to ensure core handling information is not truncated. Ledger summary needs to associate handling node data from multiple departments, requiring retrieval results to be categorized by item node or department to facilitate subsequent summary and organization.
How to configure the settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.78 | Government documents use precise terminology, so low-match results must be filtered to avoid irrelevant joint handling items interfering with retrieval |
limit | Top 10 entries | Joint handling ledgers need to associate data from multiple departments. Retrieving an appropriate number of results at one time ensures summary efficiency and avoids redundant information |
searchMode | Hybrid retrieval | Covers structured form fields and unstructured approval opinions, adapting to diverse types of joint handling documents |
usingReRank | Enabled | Re-ranks preliminary retrieval results to prioritize matching core fields such as the front-back relationship of joint handling items |
collectionIds | Specified list of department dataset IDs associated with joint handling items | Only retrieves target datasets to avoid irrelevant data from other categories mixing into retrieval results |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands query terms and supplements common terms for joint handling items to improve cross-department data recall rate |
Three common mistakes
- Phenomenon: Hybrid retrieval returns a large number of documents unrelated to joint handling items. Cause: The
similarityparameter is not set to filter low-match content, or the retrieval scope is not limited to the target joint handling dataset viacollectionIds. - Phenomenon: A query about a single PDF document returns the entire knowledge base content. Cause:
collectionIdsis not specified as the dataset to which the PDF belongs, so the retrieval scope covers all data without narrowing boundaries. - Phenomenon: Exported files do not contain the latest content after modifying the knowledge base (such as uploading a CSV file to update data). Cause: The export operation is performed before waiting for the knowledge base incremental synchronization task to complete, and data update confirmation is not completed.
How to confirm the configuration is correct
- Perform a retrieval targeting the joint handling item list, and check whether the matching scores of returned results fall within the preset 0.72–0.78 range.
- Initiate a query targeting a single PDF document, and confirm that returned results only contain content from that document, with no irrelevant information from other datasets.
- Wait for synchronization to complete after modifying the knowledge base, perform an export operation, and check that the exported file contains the latest updated content.
- After enabling hybrid retrieval and re-ranking functions, check whether returned results are sorted by the handling nodes of joint handling items, with no duplicate entries.
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.