What the data for this category looks like
The data for this category primarily comes from project document libraries, configuration lists, meeting minutes, and test reports. Daily produced documents include as-built documents, configuration manuals, training records, and acceptance certificates. Data is updated synchronously at the end of each project phase or when the overall project is completed. Documents typically include fields such as project number, version identifier, delivery milestone, and signature status. The size of individual documents varies widely; some as-built documents may contain multiple pages of technical parameters and delivery details.
What constraints these characteristics impose on knowledge base retrieval and recall
The data characteristics of this category impose multiple constraints on knowledge base retrieval and recall. Since data is updated at the end of project phases or upon overall project completion, the retrieval system must support incremental synchronization mechanisms to avoid fully reprocessing historical documents and reduce resource consumption. Documents contain precise fields such as project number and version identifier, so retrieval must match corresponding fields to prevent retrieving invalid content from cross-project or old versions. The size of individual documents varies widely, and some as-built documents include multiple pages of technical parameters, so flexible long-text chunking rules must be adapted to avoid truncating critical delivery details. Data from multiple sources is stored together, so retrieval scope must be divided by document type to enable quick location of as-built documents or acceptance certificates during pre-sales Q&A.
How to configure settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precise field requirements of project documents, filters low-relevance cross-project documents |
limit | Top 6 entries | Controls the number of recall results per query, adapts to quick filtering of valid delivery materials during pre-sales Q&A |
usingReRank | Enabled | Optimizes ranking results for long-text documents, ensures core materials such as as-built documents and acceptance certificates are retrieved first |
datasetSearchUsingExtensionQuery | Enabled | Supplements extended query terms to cover precise retrieval requirements for fields such as project number and version number |
collectionIds | Configured by current handover project group | Limits retrieval scope to the document set of the corresponding project, avoids retrieving historical data from unrelated projects |
Three common configuration mistakes
- Issue: An error prompts when importing project documents in web link format, stating that the content cannot be read. Cause: The target link requires login permissions, and no login-free proxy or import rule to skip login has been configured.
- Issue: The chat interface prompts that no knowledge base is selected, but retrieval works normally in debug preview. Cause: The document set ID of the corresponding project is not bound in the conversation chain, or front-end configurations have not taken effect synchronously.
- Issue: Retrieval results include a large number of documents not from the current project. Cause: The retrieval scope of
collectionIdsis not limited, or thesimilaritythreshold is set too low, leading to retrieval of unrelated content.
How to verify successful configuration
- Manually upload a single as-built document from the current project to trigger a retrieval test, and check whether the recall results only include relevant documents from the current project.
- Adjust the
similaritythreshold before and after to compare retrieval results, confirming that result relevance changes as expected with the threshold. - View the knowledge base synchronization log, confirming that only newly added phase or completed project documents are synchronized to the retrieval library, with no full repeated synchronization records.
- After enabling
usingReRank, verify that core delivery documents such as acceptance certificates are ranked higher than secondary documents.
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 2026-10-19.