What data looks like for this category
Data for this category comes from enterprise contract management systems, compliance risk databases, laws and regulations repositories, case management systems, and reporting platforms. Daily documents include compliance risk lists, contract legal review opinions, major issue legal opinions, compliance review forms, and violation report investigation reports. The data update frequency is monthly. A single document usually contains fields such as risk description, review conclusion, involved matters, and handling suggestions. There is no unified fixed unit, and the format varies by document type.
What constraints these characteristics impose on knowledge base retrieval and recall
Data is scattered across multiple independent business systems. Document formats and fields differ across sources. This increases text standardization processing costs during retrieval. The monthly update frequency requires the knowledge base to use periodic full-refresh synchronization logic. High-frequency incremental synchronization is not needed. Document types cover compliance risk lists, review opinions, and others. Core retrieval dimensions vary significantly across different documents. Different retrieval weights must be matched. Retrieval results directly impact core work such as compliance review and risk identification. Errors can trigger serious business risks. This creates higher relevance requirements for recalled results.
Configuration settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.85–0.92 | Legal compliance documents have high relevance requirements. This range filters irrelevant results with low semantic matching, and avoids incorrect reference to non-compliant or expired documents. |
limit | Top 8–12 results | Compliance-related searches usually require precise results. Too many results increase the screening workload for legal staff. This range covers core relevant documents. |
searchMode | Hybrid retrieval | Combines semantic and full-text retrieval. Covers compliance terms matched by keywords and risk analysis content matched by semantics. |
usingReRank | Enabled | Performs secondary ranking on initial recalled results. Raises the display priority of highly relevant documents. Meets the strict requirements of compliance review. |
datasetSearchUsingExtensionQuery | Enabled | Generates extended query terms based on user questions. Covers synonyms and relevant compliance terminology. Improves recall coverage across documents. |
collectionIds | Collection IDs divided by business module | Separates datasets for contract management, compliance risk, case management, and others. Avoids interference from irrelevant results across modules. |
Three common configuration mistakes
- Irrelevant documents with semantic similarity below 0.85 or unqualified full-text search scores appear in retrieval results. This happens when the threshold for the
similarityparameter is not configured correctly, orusingReRankfiltering is not enabled. Low-relevance content is then recalled. - The page continuously displays "Retrieving" with no returned results. This occurs when extended query terms generated after enabling
datasetSearchUsingExtensionQueryare too long, exceeding system processing limits. Or whencollectionIdsspecifies a non-existent dataset. - Retrieval results include expired compliance documents or invalidated review opinions. This happens when the knowledge base synchronization task is not configured with a monthly update frequency. Or when expired laws and regulations documents are not filtered.
How to confirm the configuration is correct
- Import a known compliance risk list document. Launch a query that includes the risk name. Verify the sorting and relevance of returned results.
- View system retrieval logs. Confirm that the filtering logic for the
similarityparameter is active. Low-score results are not returned. - Check the knowledge base synchronization task configuration. Confirm that the monthly full-refresh cycle is correctly set.
- Switch dataset IDs for different business modules. Verify that retrieval results only include documents from the corresponding module. No irrelevant cross-module content interferes.
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.