What this category of data looks like
This category of data originates from internal control systems, ERP systems, financial sharing systems, audit management systems, and procurement contract project systems of central enterprise internal control departments, audit departments, financial departments and their subordinate units. Daily generated documents include internal control self-evaluation reports, audit notices, audit working papers, problem lists, rectification ledgers, rectification closure certificates, and similar materials. The data updates on an annual cycle. Core documents contain clear problem items, rectification requirements, supporting material numbers, and closure status fields. There is no real-time high-frequency updated business flow data.
Constraints for knowledge base retrieval and recall
Low-frequency annual updates mean the knowledge base does not need high-frequency incremental synchronization. A full update strategy is suitable. Mixed structured and semi-structured documents require optimizing retrieval matching logic for problem list and rectification ledger field features. Audit documents have high evidence relevance requirements. Retrieval must prioritize matching association fields between problems and rectification supporting materials. Cross-system source data needs unified field mapping. This avoids missing fields or format incompatibility during retrieval. A single retrieval must cover historical audit and rectification data from multiple sources. A single recall result must include sufficient associated evidence items.
Configuration guidelines
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Matches relevance between audit problems and rectification supporting materials, prevents low-relevance results from entering the retrieval set |
limit | Top 10–15 items | Covers the number of associated problems and supporting materials needed for a single audit rectification report |
searchMode | hybrid | Combines keyword and semantic retrieval, adapts to structured problem lists and semi-structured rectification ledgers |
usingReRank | Enabled, re-rank top 5 results | Re-ranks recall results by rectification priority, prioritizes high-priority rectification problems |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonyms and industry terms related to audits to improve retrieval coverage |
collectionIds | Specify 3–5 business domain collection IDs | Only retrieves knowledge base collections related to internal control, audit and rectification, excludes irrelevant business data |
Three common configuration mistakes
- Phenomenon: When debugging knowledge base search, associated rectification ledger items are returned, but the AI dialogue stage does not reference knowledge base content. Cause:
datasetSearchUsingExtensionQueryis not enabled, leading to insufficient retrieval expansion, or the correct knowledge base collection ID is not bound during the dialogue stage. - Phenomenon: The number of retrieval results does not match the configured
limitparameter, with too many or too few results. Cause:collectionIdsis not set correctly, the retrieval scope includes redundant data from non-business domains, or thesimilaritythreshold is set unreasonably, causing filtering abnormalities. - Phenomenon: When processing rectification closure certificate documents with embedded images, retrievable text content cannot be generated. Cause: No image text extraction process is configured, or the URL converted from the image is not correctly associated with the knowledge base entry.
How to verify successful configuration
- View the knowledge base retrieval debugging interface, confirm that returned results include rectification ledgers and audit working papers from the specified business domain.
- Adjust the
similarityparameter, observe changes in retrieval result relevance, and verify the threshold setting meets business requirements. - Check the
collectionIdsconfiguration for the knowledge base collection, confirm only datasets related to internal control and audit are included. - Simulate an audit report generation scenario, verify that recall results cover core problems and supporting materials.
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.