What data for this category looks like
This category’s data comes from central enterprise internal control systems, ERP, financial sharing systems, audit management systems, and procurement contract project systems. The update frequency is annual. Daily documents include internal control self-evaluation reports, audit notices, audit working papers, issue lists, rectification ledgers, and rectification closure certificates. Most document structures include fields such as problem description, rectification requirements, supporting material identifiers, and closure status. Some documents include image-based supporting materials. Data sources cover multiple business systems, requiring unified integration into the knowledge base.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
The data characteristics of this category impose multiple constraints on the retrieval and recall link. Data from multiple system sources has inconsistent formats, requiring adaptation to import rules for different data sources. The annual update frequency means high-frequency incremental synchronization is unnecessary. Regular full updates can be used instead. Documents with image supporting materials require the retrieval system to support multimodal content indexing and output. The strong association between audit issues and rectification records requires precise matching of rectification ledgers corresponding to issues, to avoid mixing irrelevant content. The consequences of errors directly relate to regulatory requirements. The confidence of retrieval results must be ensured to reduce false recalls.
How to set configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale for This Setting |
|---|---|---|
similarity | 0.75–0.85 | Internal control and audit data requires precise matching of problem descriptions. This range balances recall rate and accuracy, and avoids interference from irrelevant rectification records |
limit | Top 10 entries | Audit rectification data entries are relatively standardized. 10 entries can cover typical associated rectification and issue list content for a single query |
searchMode | Semantic retrieval | Adapts to the semantic matching needs for policies and rectification requirements in party building learning Q&A, replacing the rigid matching of keyword retrieval |
usingReRank | Enabled | Audit issues are mostly structured descriptions. Reranking further filters low-relevance results and improves retrieval accuracy |
datasetSearchUsingExtensionQuery | Enabled | Party building learning Q&A requires association between policies and rectification actions. Extended queries can supplement relevant party building policy keywords for more comprehensive matching |
collectionIds | Specified by business domain | Distinguishes different datasets for internal control, audit, and rectification, to avoid mixing cross-domain irrelevant data in retrieval results |
Three common mistakes
- Symptom: The dialogue cannot output image URL content stored in the knowledge base. Cause: Retrieval output rules for multimodal data are not configured. Only text metadata is indexed, and image resource addresses are not associated.
- Symptom: When importing PPT-format audit working papers, only text is extracted and embedded images are not retained. Cause: The image extraction switch during document parsing is not enabled, or the chunk configuration does not cover image fields.
- Symptom: The same rectification issue appears multiple times independently in retrieval results, without automatic merging. Cause: The search reference merging function is not enabled, or the merging rule does not match duplicate rectification record entries.
How to confirm the configuration is complete
- Run a test query containing specific rectification issue keywords, and check that the number of returned results matches the
limitsetting. - Upload a rectification closure certificate document that includes images, and confirm that the image URL field is included in the retrieval results.
- Import a PPT-format audit notice, and check that the parsing result includes both the text within the document and embedded image content.
- Adjust the
similaritythreshold, and verify that the accuracy of retrieval results changes as expected.
The values provided 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.