What data for this use case looks like
Data sources for meeting minutes organization include full meeting recording transcripts, on-site remarks from attendees, prior meeting agenda documents, and post-meeting follow-up task ledgers. Updates are generated immediately after a single meeting completes, with each individual meeting as the independent update unit. The document structure is fixed, split into four modules: basic meeting information, agenda discussion content, resolution items, and follow-up tasks. Standardized fields include number of attendees (unit: people), discussion duration (unit: minutes), and number of task items (unit: items), among others.
What constraints these characteristics impose on knowledge base retrieval and recall
The multi-source mixed sources of meeting minutes include colloquial transcripts, which introduce redundant filler words and repeated statements. This requires the retrieval and recall link to first filter non-core information. The use of individual meetings as the update unit requires the knowledge base to support incremental synchronization by meeting dimension, avoiding performance loss caused by full re-scanning. The fixed document structure has a wide range of content lengths, from thousands to tens of thousands of characters. This requires the retrieval link to adapt to context inputs of different lengths, while prioritizing the recall of core content related to resolution items and follow-up tasks.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Filter colloquial low-relevance content in meeting minutes, retain content related to core meeting agendas and resolutions |
limit | Top 8–12 items | Adapt to the content volume of a single meeting minute, avoid context overload affecting subsequent generation |
searchMode | Full-text retrieval | Cover core information across multiple modules including meeting discussion content and resolution items |
usingReRank | Enabled | Perform secondary ranking on initially recalled results, prioritize high-value content matching resolutions and tasks |
datasetSearchUsingExtensionQuery | Disabled | The core content of meeting minutes is on-site discussion and resolutions, no need to generate additional extended queries to interfere with core information recall |
collectionIds | Configured by the knowledge base group to which the meeting belongs | Limit the retrieval scope to agenda documents and prior minutes related to the current meeting, narrow the recall range |
Three Common Misconfigurations
- Phenomenon: After the workflow is called, the returned results include the input and response fields of knowledge base retrieval, displaying the original search request and recalled document fragments. Cause: The log output configuration for retrieval results was not disabled in the workflow node, causing intermediate steps of tool calls to be exposed.
- Phenomenon: The number of knowledge base recall results far exceeds the value set for the
limitparameter. Cause: ThecollectionIdsparameter was not bound correctly, causing the retrieval scope to cover irrelevant knowledge base collections and introducing non-meeting-related content. - Phenomenon: The recall results in the rerank stage deviate from the original retrieval requirements, matching extended content unrelated to the meeting. Cause: The
datasetSearchUsingExtensionQueryparameter was enabled by default, and the generated extended queries interfered with the core retrieval logic.
How to Verify Correct Configuration
- Upload the minutes data of a single test meeting to the specified knowledge base, trigger the retrieval and recall process, and check whether the number of returned results matches the configured range of the
limitparameter. - View the output logs of the workflow node, confirm that the input and response fields of knowledge base retrieval are not included, and verify the correctness of the log output configuration.
- Check the core content of the retrieval results, confirm that entries related to resolution items and follow-up tasks are displayed first, and verify that the
usingReRankparameter is active. - Confirm that the upload limit of the locally deployed knowledge base meets the storage requirements of the current meeting minutes data, and verify the rationality of the storage space configuration.
The values given 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 2026-09-28.