What Data for This Category Looks Like
Data for this category originates from requirement specification documents, test environment configuration records, customer business sample materials, and defect system work orders. Daily produced documents include UAT use cases, acceptance issue tickets, acceptance reports, and signed confirmation sheets. Data update rhythm follows project phases or software versions. Document structures vary across different business categories. Some documents contain fields such as business scenario descriptions, test steps, problem reproduction steps, and acceptance conclusions. Content formats are adapted only to business scenarios, with no unified fixed unit requirements.
Constraints on Knowledge Base Retrieval and Recall
Data is stored by business category and sourced from multiple locations. Retrieval processes must support filtering recall results by category dimension to avoid interference from irrelevant documents. Data updated per phase or version has a large scale. Batch import and incremental update logic must be supported to prevent single-operation timeouts. Documents contain long-text test steps and problem details. Reasonable chunking logic must be implemented to ensure complete semantics during retrieval. Different documents have distinct field focus areas. Retrieval processes must balance matching between business scenarios and specific problems to improve recall accuracy.
How to Configure Settings
The following configurations are based on FastGPT v4.17.0, and adapt to knowledge base retrieval and recall requirements for user acceptance testing support scenarios.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches semantic precision for UAT use cases and acceptance issue tickets, filters low-relevance document content |
limit | Top 10 entries | Controls the number of single-recall results, adapts to the fast response scenario of operations duty and work order Q&A |
searchMode | hybrid | Combines keyword and semantic retrieval, covers matching needs for fixed test steps and vague problem descriptions |
usingReRank | Enabled | Re-ranks initial recall results, improves matching accuracy for acceptance issues and defect work orders |
datasetSearchUsingExtensionQuery | Enabled | Automatically optimizes extended queries for user questions, adapts to scenario-based expressions in customer business samples |
collectionIds | Specified by project category | Limits retrieval scope, avoids interference from cross-category irrelevant documents, adapts to phase-based filtering requirements |
Three Common Configuration Mistakes
- Importing a large number of documents in batches triggers a maxLength error. This occurs when import tasks are not split into reasonable single quantities, and the single upload limit of the knowledge base is not accommodated.
- Importing Yuque public links into the knowledge base fails. This occurs when the link is not confirmed as a directly parsable publicly available data source format, and parsing rules for non-standard web content are not accommodated.
- Retrieval results include cross-project irrelevant documents. This occurs when
collectionIdsare not specified to limit the retrieval scope, resulting in recall of document content from non-current phases.
How to Confirm Configurations Are Set Correctly
- Upload a single UAT use case document, check if the import status is normal, and confirm there are no format parsing errors.
- Input simulated acceptance questions, check if the matching scores of retrieval results fall within the preset interval, and verify recall accuracy.
- Call the retrieval API, check if the number of returned results matches the set
limitvalue, and verify the quantity control effect. - View the knowledge base category configuration, confirm the retrieval scope only includes the current project's
collectionIds, and verify the category filtering effect.
The values provided are common starting points, and should be validated against samples specific to the intended use case.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.