What this category of data looks like
Data for this category comes from the budget management integrated system, project declaration materials, fund payment data, business system ledgers, and satisfaction surveys. Updates occur annually. Daily documents include performance target forms, self-assessment forms, performance evaluation reports, supporting material lists, and rectification reports. Fields cover project ID, fund amount, performance indicators, completion status, and similar items. No unified fixed unit is used. Values match actual project conditions. Data service subjects include finance bureau performance departments, budget units, third-party evaluation institutions, and competent departments. All documents have strong project association attributes.
What constraints these characteristics impose on knowledge base retrieval and recall
The annual update feature means the knowledge base does not need frequent incremental synchronization. Full refresh must be completed within the update cycle. Mixed document formats require retrieval logic that adapts to structured table and unstructured report field extraction. Strong correlation of core project fields requires precise matching of key information such as project ID and fund amount. This avoids mixing irrelevant project data. The high rigor requirement for performance verification means recall results must meet credibility standards. This prevents score deviations caused by incorrect associations. Multi-subject data isolation rules mean retrieval scope must be strictly limited to datasets aligned with corresponding permissions.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Aligns with the field correlation of performance evaluation documents, avoids retrieving irrelevant project data |
limit | Top 8 entries | Covers the common number of supporting materials per project, ensures coverage of materials required for verification |
searchMode | Precise matching | Prioritizes matching core fields such as project ID and fund amount, meets the precise verification needs of performance evaluation |
usingReRank | Enabled | Reorders recall results by business relevance, optimizes the sorting logic for rectification reports and performance target forms |
datasetSearchUsingExtensionQuery | Disabled | This category of data is updated statically annually, no need for additional query expansion, avoids introducing irrelevant information |
collectionIds | List of dataset IDs divided by budget unit | Isolates performance data of different budget units, ensures the retrieval scope complies with permission requirements |
Three common errors
- Phenomenon: The application only returns content within the knowledge base when called. Even if the configured AI model supports online search, no external latest information is added. Cause: The association switch between model online search permission and the retrieval link is not correctly enabled in the configuration. Only private knowledge base data is forced for recall.
- Phenomenon: The number of documents returned by retrieval does not meet expectations. Results are either too many or too few. Cause: The
limitparameter value is unreasonable. It does not match the common number of supporting materials per project in this category. - Phenomenon: Irrelevant performance data across budget units appears in retrieval results. Cause: The
collectionIdsconfiguration does not limit to the dataset of the corresponding business department. The retrieval scope exceeds permitted bounds.
How to confirm the configuration is correct
- Enter the application configuration page of FastGPT v4.17.0. Check the parameters in the knowledge base retrieval module. Confirm that the values of configuration items such as
similarityandlimitmatch the preset settings. - Upload a sample performance target form for a single project. Perform a retrieval test. Check whether the returned results include the associated supporting materials and self-assessment form of the project.
- View the knowledge base call logs. Confirm that the number of documents returned by each retrieval matches the configured
limitvalue. - Test cross-departmental retrieval permissions. Confirm that configurations for different budget units can only access the content of their corresponding datasets.
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.