What data looks like for this category
This category’s data primarily comes from business systems of central enterprise internal control departments, audit departments, finance departments, subordinate enterprises, and external audit institutions. It covers documents including internal control self-assessment reports, audit notices, audit working papers, issue lists, rectification ledgers, rectification closing certificates, and more. Data sources cover enterprise internal control systems, ERP, financial sharing systems, audit management systems, and procurement contract project systems. The overall update frequency is annual. Most documents follow audit workflows: pre-event notification, in-process working papers, and post-event rectification closed-loop. Most fields include core business attributes such as issue description, rectification measures, and supporting material identifiers.
Constraints imposed on vector models and indexing by these characteristics
The annual update frequency means vector indexes do not need high-frequency incremental synchronization. Full batch update mode is suitable. Documents from multiple systems have format and field differences. Unified preprocessing logic is required to align core business fields. Documents such as rectification ledgers and rectification closing certificates mix structured and unstructured content. Indexes must support vector extraction for mixed fields. Core information of audit working papers and issue lists is concentrated in issue description and rectification requirement paragraphs. When chunking, prioritize retaining key context. Set maxToken to adapt to external API call length limits. This ensures complete vector generation for entire indexed content.
Reference only; verify against the target FastGPT release and deployment documentation.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the paragraph length of audit working papers and rectification ledgers. Retains complete context of issue descriptions and rectification requirements |
| Overlap Length | 100–150 characters | Prevents key business information across chunks from being truncated. Ensures index context consistency |
| Similarity Threshold | Calibrated based on actual testing | Adapts to the matching accuracy of audit issue keywords. Adjust based on actual business samples |
| Single Recall Limit | Top 10 entries | Covers associated samples of common audit issues. Avoids overly narrow recall ranges or excessive redundancy |
maxToken | 4096 token | Adapts to external API call length limits. Ensures complete vector generation for entire indexed content |
defaultToken | Mapped based on business fields | Unifies field alignment for documents from different systems. Ensures consistent vector extraction for unstructured content |
Three common configuration errors
- A collection is shown as created successfully, but the index status remains in initialization or fails to complete. This occurs when high-frequency incremental indexing logic is used without adapting to the category’s annual update frequency. Invalid system resource occupation follows, blocking index construction processes.
- Vector index results work normally when run locally, but all vector scores are identical after packaging as a Docker image and running. This happens when multi-system document field mapping logic is not unified. The
defaultTokenconfiguration does not take effect, causing documents from different sources to be incorrectly aligned to identical feature vectors. - Mixed retrieval request latency exceeds 10 seconds, exceeding business response expectations. This occurs when no single recall limit is set, while non-essential reranking configuration is enabled. The large retrieval scope increases system computing load.
How to confirm proper configuration
- Upload a single typical document, such as an audit working paper. Manually trigger index construction. Check whether the index progress completes within a reasonable time frame.
- Input audit issue keywords and perform retrieval. Verify that recalled results include matching issue descriptions and rectification measures.
- Compare vector generation results of documents from different system sources. Confirm that feature consistency after field alignment meets business expectations.
- Batch upload the annual updated document collection. Verify that the full index construction process completes normally.
The values provided here are common starting points. They 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.