What the Data for This Category Looks Like
This category’s data originates from requirement specification documents, test environment configurations, customer business sample materials, and defect system records. Daily produced documents include UAT test cases, acceptance issue tickets, acceptance reports, and signed confirmation sheets. All documents are updated according to test phases or version iterations. Data fields include test case steps, expected results, problem reproduction paths, defect severity levels, acceptance conclusions, and more. There are no unified mandatory units. Field content is adjusted based on the customer’s business scenario.
Constraints Imposed by These Characteristics on Re-Ranking and Hybrid Retrieval
UAT test cases and acceptance issue tickets are updated alongside project phases. Retrieval results must match the latest documents for the current acceptance phase. Hybrid retrieval must prioritize recalling recently updated data sources. Customer business sample materials have strong personalized characteristics. Hybrid retrieval must combine semantic and business scene matching. It must avoid recalling irrelevant general test documents. Defect system records have detailed problem reproduction paths. The re-ranking link must prioritize sorting results with complete reproduction steps that match current work order keywords. It must also control the number of results to fit the quick review needs of test personnel. Signed confirmation sheets are final archived documents. Retrieval must filter out historical records not matching the current version. This removes irrelevant content by version dimension.
Configuration Settings
| Configuration Item | Recommended Setting | Basis for This Setting |
|---|---|---|
similarity | 0.65–0.75 | Business semantic matching for UAT test cases must balance recall rate and precision. This range filters low-relevance general test documents and fits retrieval needs for acceptance scenarios. |
limit | Top 8–12 entries | Test personnel need to quickly review results. Too many entries increase screening costs. This range fits the effective information capacity of a single retrieval round and aligns with daily review habits. |
searchMode | Hybrid retrieval mode | Must combine keyword matching and semantic recall. This covers step descriptions in UAT test cases and keyword tags in acceptance issue tickets, fitting retrieval needs for multiple document types. |
usingReRank | Enabled | Must re-rank initial recall results by business relevance. Prioritize displaying defect records with complete reproduction paths and matching test cases. This improves efficiency of problem reproduction and classification. |
| Re-ranking model call threshold | Calibrated via actual testing | UAT document characteristics vary widely across different projects. Adjust the threshold through actual retrieval results to match the needs of specific business scenarios. |
| Version filter toggle | Enabled | Only recall documents from the current acceptance phase. Avoid interference from historical version UAT test cases on retrieval results. This meets timeliness requirements for project acceptance. |
Three Common Configuration Mistakes
- Phenomenon: Retrieval results show no obvious changes after enabling
usingReRank. Cause: No valid API key for the re-ranking model is associated, or the input text length of the re-ranking model exceeds the interface limit. This causes the re-ranking link to not trigger normally. - Phenomenon: Hybrid retrieval returns a large number of UAT test cases from non-current phases. Cause: No filtering rules configured by project version or phase. This recalls documents from past acceptance phases, interfering with current work order processing.
- Phenomenon: The number of retrieval results does not match expectations after setting
similarity. Cause: Confused the functional logic ofsimilarityandlimit. Mistakenly used the similarity threshold as a restriction on the number of results. This leads to retrieval results exceeding or falling short of expected quantities.
How to Confirm Configuration Is Correct
- Manually enter keywords from an acceptance issue ticket. Check if initial retrieval results include relevant UAT test cases and defect system records.
- Enable
usingReRank, then compare the result order before and after re-ranking. Confirm that documents with high-match reproduction paths appear at the top. - Adjust the value of
similarity, observe changes in the number of retrieval results. Confirm the configuration takes effect. - Check if retrieval results only include documents from the current project phase. Confirm the version filter toggle is configured correctly.
This verification process applies to FastGPT v4.17.0.
Note: All provided configuration values are common starting points. Validate them against your own sample data to match specific business requirements.
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.