What the data for this category consists of
Data for this category primarily comes from enterprise internal control systems, ERP, financial sharing systems, audit management systems, and procurement contract project systems. Daily processed documents include internal control self-assessment reports, audit notices, audit working papers, issue lists, rectification ledgers, and rectification closing certificates. The most time-consuming daily tasks are extracting business samples and vouchers, compiling audit issues and rectification requirements, and tracking rectification supporting materials and closing. Data update frequency follows an annual cycle. Most documents are a mix of structured and semi-structured formats, with fields including issue number, rectification responsible person, completion deadline, and supporting material number. The length of individual documents varies widely.
What constraints these characteristics impose on knowledge base retrieval and recall
Decentralized data sources lead to inconsistent document formats across systems. This requires adapting parsing and import workflows for multi-source data. Annual update frequency means the knowledge base does not need frequent incremental synchronization. However, single batch import data volume is large. This requires configuring reasonable chunking and indexing strategies. Documents include structured fields such as issue number, rectification responsible person, and completion deadline. Retrieval must support precise matching of business identifiers. Individual document length varies widely. Chunking must balance context integrity and retrieval accuracy. Avoid chunking too finely to lose business associations, or too broadly to cause redundant recall. Audit errors can lead to repeated audit issues, regulatory notifications, liability investigations, or financial statement adjustments. Retrieval result accuracy directly affects work quality.
How to configure the settings
These configurations apply to FastGPT v4.17.0.
| Configuration | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.85 | Matches the specialized terminology requirements of audit documents, avoids recalling irrelevant rectification records or self-assessment reports |
limit | Top 10 entries | Controls the number of single-recall results, aligns with the work habit of auditors quickly browsing content, prevents information overload |
searchMode | local | Prioritizes matching structured business fields in the local knowledge base, ensures consistency between retrieval results and internal enterprise data |
usingReRank | Enabled | Performs secondary sorting on recall results, improves matching accuracy with audit issues and rectification requirements |
datasetSearchUsingExtensionQuery | Enabled | Automatically generates extended query terms based on user questions, covers audit-related synonyms and business abbreviations, such as matching "rectification closing" to "rectification completion certificate" |
collectionIds | Specified by business module | Only retrieves dedicated collections such as internal control self-assessment, audit working papers, rectification ledgers, prevents mixing document data from other categories |
Three common mistakes
- Retrieval returns the
text index required for $text queryerror. This occurs when the knowledge base text index build process is not completed, or the corresponding index rule is not enabled during parameter configuration. - After migrating the knowledge base to a local deployment environment, retrieval results include a large number of irrelevant documents. This occurs when the
collectionIdsparameter is not used to specify the dedicated retrieval collection, causing documents from other categories to be included in the retrieval scope. - Single batch retrieval takes more than 60 seconds. This occurs when the
limitparameter is not set reasonably, andusingReRankis not enabled, causing the system to process excessive redundant data.
How to confirm the configuration is complete
- Upload a single audit working paper, run a retrieval test, and verify that returned results only include rectification ledgers and issue lists related to this paper.
- Check the knowledge base index status, confirm that all collections specified by
collectionIdshave completed text index construction. - Simulate an audit query, check that retrieval results include structured field matches such as issue number and rectification responsible person.
- Test the batch import process, confirm that no
text index required for $text queryerrors occur.
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.