What this category of data looks like
This category of data is sourced from requirement specification documents, test environment operation logs, customer business sample data, and defect management systems. Documents produced daily include UAT use cases, acceptance issue tickets, acceptance reports, signed confirmation documents, and similar materials. Data updates are triggered per project phase or software version. Data fields include business scenario descriptions, test steps, expected results, problem reproduction paths, acceptance conclusions, and other items. There are no unified mandatory units. The data adapts to the formatting of business documents.
Constraints imposed on the knowledge base retrieval and recall link
The phase or version-level update rhythm of this category of data requires the knowledge base retrieval and recall link to support switching associated datasets based on project phase or software version. This prevents redundant recall across versions. Documents produced daily mainly consist of structured test cases and issue tickets. Retrieval must accurately match keywords such as business scenarios and test steps. This avoids irrelevant results interfering with acceptance judgments. This category of data draws from customer business sample data. Recall results must support adaptation to business language, and sensitive information leakage must be avoided. Data fields change flexibly based on business document formats. The recall logic must support dynamic matching of document field structures.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 match threshold | Adapts to keyword matching accuracy for structured test cases, filters low-relevance acceptance documents |
limit | Top 10 entries | Matches the processing scale of single-batch UAT use cases, controls the context length of retrieval results |
searchMode | vector | Prioritizes matching semantically related test scenarios, adapts to queries about problem reproduction scenarios for acceptance issues |
usingReRank | Enabled, re-rank top 5 results | Performs secondary sorting on recall results, improves display priority of highly relevant acceptance issue tickets |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands synonyms of business questions, adapts to colloquial queries about test scenarios |
collectionIds | Bind corresponding datasets by project phase | Matches the phase-level data update rhythm of this category, avoids cross-phase data confusion |
Three common mistakes
- After upgrading to FastGPT v4.17.0, an error occurs when creating a knowledge base using a URL. Background logs show
cannot fetch internal url. The cause is that the internal URL of the test environment has not been added to the system access whitelist. - After uploading a Word-format UAT use case that contains images, retrieval queries do not return image content or related prompts. The cause is that the image extraction rule for document parsing has not been configured.
- A locally deployed knowledge base only supports 30 datasets, and this cannot be adjusted through conventional configuration. The cause is that the system default dataset quantity limit parameter has not been modified.
How to confirm configurations are properly set
- Execute a single business scenario query, verify that the matching degree of returned results falls within the preset interval of the
similarityparameter. - Check the number of retrieval results, confirm that it matches the value of the
limitparameter. - Upload a UAT use case that contains images, verify whether image-related content is returned after retrieval.
- Switch the dataset binding for the project phase, confirm that retrieval results only include acceptance documents from the current phase.
The values provided here are common starting points, and should be evaluated against samples from the relevant deployment environment.
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.