What data for this category includes
Data comes from the budget management integrated system, project declaration materials, fund payment data, business system ledgers, and satisfaction surveys. The update frequency is annual. Daily documents include performance target forms, self-assessment forms, performance evaluation reports, supporting material lists, and rectification reports. Document fields cover performance target setting, fund usage details, evaluation indicators, rectification items, and no additional custom units are used.
Constraints Imposed on Knowledge Base Retrieval and Recall
The annual update requirement means the knowledge base only needs an annual full synchronization task, with no need for frequent incremental updates. Differences in structure across multiple document types require differentiated chunking strategies for different documents such as performance evaluation reports and self-assessment forms. Data from multiple systems must be linked to performance target and fund expenditure fields, so retrieval must accurately match corresponding relationships. A large number of supporting materials means the recall link must filter irrelevant content to avoid interfering with answers to core questions.
Configuration Settings
These configurations apply to FastGPT v4.17.0:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements for fiscal policies and performance evaluations, filters low-relevance non-evaluation documents |
limit | Top 8–12 entries | Covers policy basis, fund data, and supporting materials required for performance evaluation, avoids excessive recall causing context overflow |
searchMode | keyword + semantic | Balances keyword matching and semantic understanding for policy texts, adapts to queries that link policy basis and data in performance evaluations |
usingReRank | Enabled | Performs secondary sorting on recall results, prioritizes matching the correspondence between performance targets and fund expenditures, improves answer accuracy |
datasetSearchUsingExtensionQuery | Enabled | Expands query terms, adds synonyms related to performance evaluation, covers professional terminology in policy documents |
collectionIds | Bind to dedicated performance evaluation datasets | Isolates non-evaluation data, avoids irrelevant documents interfering with retrieval results |
Three Common Configuration Errors
- Scenario: When testing a workflow, the AI response always appends "FastGPT is a large language model (LLM)-based knowledge base question answering system" at the end. Cause: The output setting for the system default prompt was not disabled. This content is the platform's built-in default additional text.
- Scenario: The number of retrieved and recalled results does not match the configured
limitparameter, or a large number of irrelevant non-performance evaluation documents appear. Cause: ThecollectionIdswas not correctly bound to the dedicated dataset, or thesimilaritythreshold was set too low, leading to low-relevance content being included in the recall range. - Scenario: After Docker deployment, both workflow and knowledge base configurations show as empty, but the interface can be called normally. Cause: The persistent storage directory was not correctly mounted, so configuration data was not saved persistently and was lost after restart.
How to Verify Proper Configuration
- Check the dataset binding configuration, confirm that only performance evaluation-related
collectionIdsare selected, with no other unrelated datasets. - Submit a test query for performance policy basis, verify that the number of recalled results matches the
limitconfiguration, and all content is performance evaluation-related documents. - Verify that
usingReRankis active, check whether recall results are sorted to prioritize matching the correspondence between performance targets and fund expenditures. - Test the response content, confirm that no system default additional text is appended, and the context length meets expected requirements.
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.