What This Category of Data Looks Like
This category of data primarily comes from the unified budget management system, project declaration materials, fund payment data, business system ledgers, and satisfaction surveys. The update frequency is annual. Commonly processed documents include performance target forms, self-evaluation forms, performance evaluation reports, supporting material lists, and rectification reports. Document fields include project number, fund amount, performance indicators, completion status, supporting material association items, and other items. There is no fixed unified unit, and matching is done according to actual project conditions.
What Constraints These Characteristics Impose on the Knowledge Base Retrieval and Recall Link
The annual update feature means knowledge base data does not require frequent refreshing. A single recall must cover all associated information for full-year projects. The combination of multiple document types and structured fields requires retrieval processes to balance semantic matching and precise field matching. This prevents irrelevant content from being included in generalized retrieval. Retrieval processes must integrate cross-source associated information. This ensures the corresponding relationship between performance targets, fund expenditures, and supporting materials is fully recalled. The accuracy of retrieval results directly affects performance verification and report writing. Incorrect recalls lead to verification deviations in subsequent links.
How to Configure Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.6–0.7 | High association requirements for documents in this category, filter low-relevance content to avoid mixing performance data from non-corresponding projects |
limit | Top 10 | The number of annual projects is large, limit the number of recalled entries to avoid context overload, while covering performance information for core projects |
searchMode | vector+keyword | Balance semantic matching and structured field matching, adapt to precise retrieval needs such as project numbers and fund amounts in performance evaluation |
usingReRank | Enabled | Re-sort after multi-document recall, ensure content with the highest association with performance targets and fund expenditures is returned first |
datasetSearchUsingExtensionQuery | Enabled | Expand retrieval scope to cover indirectly associated supporting materials in performance reports, meet the need to verify the corresponding relationship between funds and performance targets |
collectionIds | Bind corresponding knowledge base collections by business domain | Isolate performance data from different competent departments and budget units to avoid cross-subject data confusion |
Three Common Misconfigurations
- Symptom: Retrieval results do not include the self-evaluation form content of a specified project, and the number of returned results is insufficient. Cause: The
similaritythreshold is set too high, filtering out some documents that meet the required association level, or thelimitvalue is too small to cover all relevant projects. - Symptom: The AI answer does not display the source of the answer, making it impossible to trace supporting materials. Cause: The source association logic for knowledge base retrieval is not configured correctly, or the dedicated knowledge base collection containing supporting materials is not bound.
- Symptom: Retrieval results include performance data from other budget units, with cross-subject irrelevant content. Cause: Collections are not isolated via the
collectionIdsconfiguration, leading to cross-domain data being recalled.
How to Confirm Proper Configuration
- Manually enter the performance target number of a specific project, check whether the retrieval results include the self-evaluation form, fund payment data, and corresponding supporting materials of that project.
- View the source field of the retrieval return results, confirm that each result is associated with the corresponding document name and source location.
- Switch query keywords for different budget units, check that the recalled results only include performance data from the corresponding units, with no cross-subject content.
- Test enabling and disabling
usingReRank, compare the sorting rationality of retrieval results, and confirm that the re-ranking function is active.
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.