What the data for this category looks like
Data is derived from follow-up summaries automatically generated from hotline agent call transcripts, and follow-up record fields associated with tickets. It is automatically generated and synced to the knowledge base by the customer service system immediately after a single follow-up is completed. There is no fixed batch update cycle. Documents use a structured format with fixed fields including follow-up recipient ID, follow-up time, follow-up reason, customer feedback content, key agent responses, pending follow-up items, and more. Each document focuses on a single follow-up scenario, with content centered on specific communication details and follow-up actions.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The structured fixed field feature requires retrieval to target specific fields for targeted recall, avoiding interference from irrelevant fields on retrieval results. The real-time update feature requires the knowledge base synchronization mechanism to match the time node of follow-up completion, ensuring recalled content is the newly generated follow-up result. The feature that each document focuses on single communication details requires retrieved recall results to be strongly relevant to the current ticket's follow-up scenario, avoiding recall of cross-scenario follow-up content. At the same time, follow-up record summary content relies on the synchronization interface of the front-end customer service system. The response speed of the retrieval link must adapt to the real-time requirements of the hotline scenario, avoiding excessive waiting time.
How to configure settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Follow-up record summaries focus on specific communication details. This range balances precision and recall coverage, filtering low-relevance results while retaining valid recall content |
limit | Top 3–5 entries | A single follow-up record summary has focused content. Excessive recall will exceed the context window limit. This quantity covers recent key follow-up information |
searchMode | Semantic search | Follow-up record summaries contain a large amount of colloquial communication details. Semantic search can match implicit customer needs and key agent responses, which is better than keyword matching |
usingReRank | Enabled | Can perform secondary ranking on initial recall results, filtering follow-up records that are semantically similar but do not match the scenario, improving result relevance |
datasetSearchUsingExtensionQuery | Calibrate based on actual testing | Follow-up record summaries have strong scenario specificity. Whether to enable this should be adjusted based on the specific association rules of the ticket, avoiding introducing irrelevant knowledge base content |
collectionIds | Bind only the dedicated follow-up record summary knowledge base collection | Avoid retrieving knowledge base content from other categories, ensuring recall results only come from the target follow-up record summary data |
Three common misconfigurations
- Phenomenon: Non-follow-up record summary documents are mixed into retrieval results. Cause:
collectionIdsis not configured correctly, and a knowledge base collection containing other categories of data is bound, leading to cross-category retrieval. - Phenomenon: Corresponding follow-up records cannot be associated with ticket variables during retrieval. Cause: Variable binding rules are not configured correctly, and the follow-up recipient ID of the ticket is not associated and matched with the follow-up recipient ID field of the follow-up record summary.
- Phenomenon: The number of results returned by a single retrieval is far lower than expected. Cause: The
similarityvalue is set too high, filtering some follow-up record content relevant to the current ticket scenario.
How to verify correct configuration
- Manually submit a simulated follow-up record summary, perform a retrieval action, and check whether the returned results only contain follow-up record summary content.
- Adjust the configuration item values, observe changes in the relevance of retrieval results, and confirm that the values meet the needs of the current scenario.
- Check the knowledge base synchronization logs to confirm that newly generated follow-up record summaries are synced to the knowledge base collection within the preset time.
- Bind ticket variables to perform retrieval, and check whether the returned results are associated with records of the corresponding follow-up recipient.
Note: The configuration values listed above are common starting points. They should be evaluated against the samples relevant to the deployment context.
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.