What Data for This Category Looks Like
Data for this category primarily comes from point tracking systems, data warehouses, CRM systems, and ticket systems. Update frequency is daily or weekly. Regularly produced documents include usage analysis reports, point tracking plans, indicator caliber tables, and dashboard requirement forms. Most documents use a combination of structured fields, including indicator name, event code, statistical caliber, business definition, and other fields. Common units are times, person-times, percentages, and other general business units. Core related subjects include product operations, data analysts, and customer success operations. Content focuses on product function usage and customer behavior. This category addresses business issues such as unified indicator calibers and abnormal data troubleshooting.
Constraints Imposed by These Characteristics on Knowledge Base Retrieval and Recall
Data sources are scattered and linked to multiple types of business systems. Retrieval must be limited to datasets exclusive to product operations, to avoid mixing irrelevant content such as customer tickets or sales data. Documents primarily focus on structured indicators and caliber explanations. Retrieval must precisely match core fields such as indicator names and business definitions, rather than relying solely on literal keyword matching. Update frequency is daily or weekly. Recall results must link to the latest caliber tables and point tracking plans, to avoid using expired indicator definitions. Additionally, user questions for this category often involve indicator aliases and custom calibers. Retrieval must support extended queries to cover more related caliber explanation documents, improving retrieval coverage and accuracy.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Indicator caliber documents for product usage data require high matching accuracy, to avoid mixing irrelevant point tracking descriptions or business explanations in results |
limit | Top 8–12 entries | Documents in this category are mostly structured indicator explanations with short individual content. Excessive results increase context processing overhead |
usingReRank | Enabled | The association relationships of product usage data are complex. Re-ranking models can filter results that only match literally but are logically irrelevant, such as confusing point tracking indicators across different business lines |
datasetSearchUsingExtensionQuery | Enabled | User questions for this category often include indicator aliases or custom calibers. Extended queries can cover more related caliber explanation documents |
searchMode | hybrid | Combines semantic matching and keyword matching to precisely recall documents containing indicator names and business definitions, adapting to the retrieval needs of structured documents in this category |
collectionIds | Pass the collection ID of the knowledge base exclusive to product usage analysis | Limit retrieval scope to avoid mixing business documents from other categories |
Three Common Misconfigurations
- Phenomenon: When calling the API, the
usingReRankparameter is configured as enabled, but the re-ranked result field in the returned response is false. Cause: The deployment permission of the re-ranking model is not bound to the corresponding knowledge base collection, or the deployment address of the re-ranking model is not configured on the FastGPT platform. - Phenomenon: There is a significant difference between the retrieval results of online chat and API calls, and the accuracy of online chat is higher. Cause: Online chat enables
datasetSearchUsingExtensionQueryandusingReRankby default, but API calls do not explicitly configure these parameters, leading to inconsistent retrieval scope and matching logic. - Phenomenon: The retrieval results contain a large number of irrelevant customer ticket contents, and the target indicator caliber documents are not matched. Cause: The
collectionIdsparameter is not limited to the knowledge base collection exclusive to product usage analysis, and documents from other business categories are mixed in, expanding the retrieval scope.
How to Verify Successful Configuration
- View the knowledge base configuration page of FastGPT v4.17.0, confirm that the
collectionIdsparameter only includes datasets related to product usage analysis. - Initiate a test query containing indicator aliases, confirm that after enabling
datasetSearchUsingExtensionQuery, the corresponding indicator caliber explanation documents can be retrieved. - Call the API and pass
usingReRankas true, check if the returned response contains re-ranked sorting fields, and the sorting follows business logic, such as ranking the most relevant indicator caliber documents first. - Compare the retrieval results of online chat and API calls. Under the premise of consistent configuration, the difference between the two results should be within a reasonable range, with no significant deviation.
Note: The values provided here are common starting points and should be evaluated 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.