What data in this category looks like
This category of data comes mainly from local public data open platforms, integrated big data platforms, departmental business databases, and electronic certificate libraries. Updates occur monthly. Daily maintenance includes documents such as public data resource catalogs, data open lists, data item descriptions, desensitization rules, interface documents, and update frequency descriptions. Most data takes the form of structured metadata, including fields like data item name, open conditions, interface address, desensitization rules, and update cycle. Some documents include interface call instructions and permission boundary instructions.
What constraints do these characteristics impose on the "knowledge base retrieval and recall" link
The monthly update cycle requires the knowledge base to use a matching synchronization frequency to avoid inconsistencies between catalog information and publicly available actual data. The predominantly structured data structure requires the retrieval process to balance precise field matching and full-text semantic matching, ensuring that key metadata such as interface addresses and open conditions can be retrieved. The data involves departmental permission boundaries and desensitization rules, so retrieval must limit scope by department or data type to prevent cross-permission retrieval of sensitive data. Data accuracy directly affects the quality of service inquiry responses, so retrieval precision and the integrity of retrieved fragments must be strictly controlled to avoid incorrect guidance.
How to set the configuration
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | High precision is required for public data catalog field matching to avoid retrieving irrelevant open data items |
limit | Top 10 | Responses to service inquiries need to be concise and clear; too many entries will increase the information reception burden |
searchMode | Hybrid retrieval | The data includes structured fields such as interface address and open conditions, and unstructured text such as desensitization rule descriptions. Hybrid retrieval can balance precise matching and semantic matching |
usingReRank | Enabled | There are a large number of public data documents, and re-ranking can improve the sorting priority of highly relevant fragments and avoid key information being buried |
datasetSearchUsingExtensionQuery | Enabled | Inquiries about services often include vague expressions, and extended queries can cover more relevant catalog entries and interface descriptions |
collectionIds | List of collection IDs divided by department or data type | Public data is managed by department, and dividing by collections can avoid ineffective cross-department permission retrieval and ensure retrieved data conforms to the open scope |
Three common mistakes
- Phenomenon: A "no permission to access" prompt is returned during retrieval, or unauthorized sensitive data fields are retrieved. Cause: The retrieval scope is not limited via
collectionIds, resulting in cross-department or non-open data being retrieved. - Phenomenon: When searching interface documents, the results do not include interface address or update frequency fields. Cause:
datasetSearchUsingExtensionQueryis not enabled, so the implicit structured field requirements in the query cannot be matched. - Phenomenon: Retrieval results are sorted chaotically, with highly relevant catalog entries ranked after low-relevant entries. Cause:
usingReRankis not enabled, and thesimilaritythreshold is set incorrectly, resulting in failure of the sorting logic.
How to confirm the configuration is complete
- Enter the FastGPT knowledge base management page, check whether the collections configured in
collectionIdsare divided by department or data type. - Initiate a test query, such as "Query the open conditions of medical insurance data", confirm that the retrieved results include data item descriptions and open condition fields.
- View the retrieval log, confirm that
usingReRankanddatasetSearchUsingExtensionQueryare in effect, and the returned results are sorted as expected. - Trigger the monthly synchronization task, confirm that the knowledge base update log is normal, and there are no errors about data inconsistency.
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 2026-09-28.