What this category of data looks like
Daily documents for this category typically include joint handling item lists, one-time notification sheets, joint handling application forms, material circulation forms, parallel approval opinions, and completion receipts. Data sources include the "One Thing" special zone of the national integrated government service platform, local integrated government service platforms, electronic certificate repositories, and departmental approval systems. Documents are updated daily. Most documents are categorized by joint handling items, and include fields such as cross-departmental item association rules, material reuse requirements, and parallel processing node requirements. There is no standard fixed format, and content adjusts based on local government service specifications.
What constraints do these characteristics impose on the knowledge base retrieval and recall workflow
Decentralized data sources and inconsistent formats cause missing fields and semantic deviation during retrieval. The daily update frequency requires the knowledge base synchronization pipeline to support incremental updates, to avoid data lag. Joint handling items include cross-departmental association rules. A single document’s semantic fragment cannot fully cover the full handling logic, leading to fragmented recall results. Content related to material reuse is scattered across multiple documents. Precise matching of the reuse scenario corresponding to the user’s question is required; otherwise, recommended materials will be incomplete.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Semantic similarity threshold for matching joint handling policy text. This range filters low-relevance cross-departmental irrelevant content while retaining weakly matched results across item associations |
limit | Top 12 entries | Joint handling items involve multi-department documents. A single recall of this number is the minimum count sufficient to cover association rules and material requirements, avoiding context overload |
searchMode | hybrid | Combines keyword and semantic retrieval, adapting to the content structure of joint handling documents that have both clear rule entries and vague association descriptions |
usingReRank | Enabled | Re-sorts recall results by relevance to the user’s question, resolving sorting deviations caused by semantically scattered cross-departmental documents |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonymous expressions of user questions, covering localized expression differences common in joint handling items |
collectionIds | Dataset IDs bound by joint handling item category | Limits the retrieval scope to only government documents corresponding to the current joint handling item, avoiding interference from irrelevant data |
Three Common Misconfigurations
- Recall results fail to cover complete joint handling item association rules, with some key content missing. Cause: Contextual associations of cross-departmental rules are not preserved during text chunking, leading to key semantic fragments being split. Only partial fragments are retrieved during searches.
- A large number of individual approval documents unrelated to the current joint handling item appear in retrieval results. Cause: The retrieval dataset scope is not limited via
collectionIds, leading to accidental recall of unrelated government documents. - When a user submits a question using localized expressions, matching corresponding policy content fails. Cause:
datasetSearchUsingExtensionQueryis not enabled. The system cannot automatically expand synonymous expressions for the user’s question, leading to semantic matching failure.
How to Confirm Configurations Are Effectively Set
- Upload a test document of a joint handling item list, initiate a corresponding question query, and verify that recall results only include content related to the target joint handling item to confirm the
collectionIdsconfiguration is effective. - Adjust the value range of
similarity, verify changes in recall result relevance, and confirm the threshold setting meets business requirements. - Initiate a query using localized expressions, such as replacing "joint handling" with "multi-department joint handling", and verify that correct content is recalled to confirm the
datasetSearchUsingExtensionQueryconfiguration is effective. - Check the sorting of retrieval results, confirm that highly relevant documents are displayed first, and verify the
usingReRankconfiguration is effective.
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.