What data looks like for this category
The data for this category primarily comes from Git repositories, CI pipelines, static scanning reports, and defect databases. Daily documents include merge requests, code review comments, commit records, and branch policy descriptions. Each entry includes fields such as change scope, review status, associated defects, and submitter information. Data updates occur either when commits are made or on a daily synchronization schedule. A single commit generates new review records and merge request entries. Branch policy descriptions are updated in line with adjustments to team rules.
What constraints do these characteristics impose on knowledge base retrieval and recall
Data for this category centers on code changes and review records. Most fields involve technical details and associated business, with a relatively high update frequency. In operations on-call scenarios, users need to quickly locate historical merge requests and review comments related to the current ticket. Retrieval accuracy directly affects on-call efficiency. This requires the knowledge base retrieval to accurately match code change scopes and review contexts, avoiding irrelevant non-code document recalls. The frequent update rhythm requires real-time or near-real-time synchronization mechanisms to ensure recalled content aligns with the latest merge requests and review statuses. Additionally, fields for associated defects and submitter information support filtering by associated dimensions during retrieval, narrowing the recall scope and improving query efficiency during operations on-call. The high semantic complexity of code documents also requires the retrieval model to accurately identify technical terms and change logic, preventing incorrect recall of unrelated content.
How to configure settings
This document is written based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches semantic similarity between code changes and review comments. This range filters low-relevance non-target review records while retaining valid matching results |
limit | Top 8 entries | Controls the number of merge requests and review records recalled per single query, adapting to the quick browsing needs of operations on-call scenarios and avoiding excessive results that interfere with judgment |
searchMode | Semantic retrieval | Matches technical semantics of code changes and review comments, can recall implicitly associated review content, and performs better than basic keyword retrieval |
usingReRank | Enabled | Performs re-ranking on initial recall results, prioritizing displaying merge requests or review records with higher relevance to the current operations ticket |
datasetSearchUsingExtensionQuery | Enabled | Generates extended semantics based on user queries, covering scenarios such as synonyms for code changes and branch naming conventions, improving recall coverage |
Three common configuration errors
- Search results are empty or fail to match target merge requests. The cause is that the
similaritythreshold is set too high, filtering some semantically relevant review records that do not meet the similarity standard, orsearchModeis not correctly configured as semantic retrieval mode. - The number of knowledge base recall results does not meet expectations, with the number of single returns exceeding the configured range. The cause is that the
limitparameter setting does not match the browsing needs of the operations scenario, or the post-reranking results are not properly truncated. - Training fails for some merge request files after upload, and content cannot be read during recall. The cause is that correct chunking rules are not configured for code files, or binary files are not skipped during upload, leading to training failure.
How to verify successful configuration
- Enter the knowledge base search test interface, enter a typical operations ticket query, and verify that the number of returned results matches the
limitconfiguration. - Adjust the
similaritythreshold, observe changes in the relevance of returned results, and confirm that the threshold range meets business requirements. - After enabling
usingReRank, compare the order of initial recall results and post-reranking results to confirm that the reranking function works correctly. - Upload a test merge request file, trigger synchronization, and perform a search to confirm that the content of this file can be correctly recalled.
The values provided here 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 2026-10-19.