What this category of data looks like
Data for this category mainly comes from the unified budget management system, department staff establishment system, prior-year final accounts, project library, and the Ministry of Finance’s budget preparation notices. Daily documents include department budget drafts, project application forms, performance target forms, personnel expense calculation forms, budget approvals, and similar materials. These documents typically include fields such as project codes, budget amounts, and staff establishment numbers. The update frequency is annual, with all updates and circulation completed only during the annual budget cycle. Involved entities are mostly departments such as the budget office of the finance bureau and the finance office of budget units.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The annual update feature means the knowledge base only needs a full refresh during the annual budget preparation phase, without frequent incremental synchronization. This reduces daily operation and maintenance costs. The mix of multiple types of structured documents requires the retrieval link to configure differentiated field matching weights for different document types. This ensures that core business fields such as budget amounts and project codes have higher recall priority. The need to integrate content across data sources requires retrieval and recall to cover both party building learning policies and budget preparation specifications. Cross-dataset retrieval capability must be configured to avoid limiting the retrieval scope to a single document type. Business errors directly affect fund allocation, so the accuracy of recall results must be strictly controlled to avoid irrelevant content interfering with the review process.
How to configure the settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the text accuracy of structured budget preparation fields and party building policies, balancing recall rate and precision |
limit | Top 10 entries | Adapts to the volume of annual budget documents, avoiding context overload from excessive recall results |
searchMode | Hybrid retrieval | Supports both structured field matching and unstructured text recall, covering both budget drafts and party building policy content |
usingReRank | Enabled | Performs re-ranking on initial recall results to improve matching relevance between party building policies and budget preparation requirements |
datasetSearchUsingExtensionQuery | Enabled | Optimizes queries using historical conversations to resolve reference ambiguity in party building contexts |
collectionIds | Predefined collection IDs for budget preparation document library and party building policy library | Accurately specifies the retrieval scope, avoiding interference from unrelated dataset content |
Three common misconfigurations
- Issue: When calling the knowledge base API, retrieval results do not incorporate historical conversation optimization, leading to ambiguous references. Cause: The
datasetSearchUsingExtensionQueryconfiguration is not enabled, and the historical conversation extended query function is not activated. - Issue: After the knowledge base retrieval is triggered, GPU resources show no usage, and the retrieval task times out without returning results. Cause: The dataset storage node corresponding to
collectionIdsis not correctly associated, causing the retrieval request to fail to be scheduled to GPU computing resources. - Issue: After the retrieval configuration is saved, restarting the FastGPT v4.17.0 service automatically restores the default parameters. Cause: Configuration persistent storage is not enabled, or cluster node synchronization is not completed when the configuration is submitted.
How to verify the configuration is complete
- Access the FastGPT knowledge base management interface, check whether the
collectionIdsconfiguration is bound to the corresponding collection IDs of the budget preparation document library and party building policy library. - Initiate a test retrieval, input a combined query containing party building policy keywords and budget preparation terminology, and verify whether the returned results include both types of content and follow the expected sorting order.
- Save the configuration, restart the service, re-enter the configuration interface, and confirm that the parameters have not been restored to default values.
- Call the knowledge base API, pass historical conversation context containing ambiguous references, and verify whether the returned results have resolved the reference ambiguity issue.
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.