What Data for This Category Looks Like
Core data for this category is maintained by data service project managers, collection teams, and client algorithm teams. Daily documents fall into four categories: collection plans, sample specifications, authorization instructions, and task schedules. Data sources include client requirement documents, public data source lists, collection tool configuration records, and internal sample libraries. Data update frequency adjusts with project progress or weekly. Documents within a single project update dynamically as sample scope and field specifications are confirmed. Document fields include professional content such as sample coverage, data format requirements, and collection legality verification items. There is no unified general template, and customization is required per individual project requirements.
Constraints on Knowledge Base Retrieval and Recall
Documents for data collection task design are managed independently per project and updated frequently. Retrieval must be limited to the dataset scope of a single project to avoid mixing data across projects. Documents contain sensitive content such as professional field specifications and legality verification items. High matching precision is required to ensure recalled content accurately matches the specific needs of user inquiries. The number of documents within a single project is moderate, but fields are highly professional. Prioritize recalling content strongly related to the current inquiry to avoid redundant results interfering with decision-making. Additionally, user inquiries often include unstated needs such as collection legality and sample scope confirmation. Extended query support is required to cover these implicit contents.
Configuration Settings
The following configurations are formulated based on the rules of FastGPT v4.17.0.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
collectionIds | Bind the dataset ID set corresponding to the current project | Documents for data collection task design are managed independently per project. Retrieval scope must be limited to avoid mixing data across projects |
similarity | 0.75–0.85 | Collection task documents include professional field specifications and legality instructions. High matching precision is required to ensure recalled content is accurate |
limit | Top 6–10 items, calibrated via actual testing | The number of collection task documents within a single project is moderate. Return an appropriate number of results to cover core content such as plans, schedules, and specifications |
usingReRank | Enabled | Recalled results must be sorted by relevance to prioritize matching key inquiry content such as sample scope and field definitions |
datasetSearchUsingExtensionQuery | Enabled | Extended queries can be generated based on user questions to cover implicit detailed needs in collection plans |
searchMode | Semantic retrieval | Adapts to the unstructured text characteristics of collection task documents and accurately matches professional terminology |
Three Common Configuration Mistakes
- Leaked customer authorization information in retrieval results. The symptom is that returned results include collection authorization instruction documents from other customers. The cause is incorrect configuration of
collectionIds, leading to retrieval scope covering datasets outside the current project. - Redundant and poorly relevant recalled results. The symptom is returning more than 10 irrelevant collection schedule documents. The cause is a
similarityvalue lower than 0.75, with matching precision failing to meet the retrieval requirements for professional documents. - Failure to cover implicit inquiry needs. The symptom is that authorization instruction documents are not recalled when a user asks about collection legality. The cause is that
datasetSearchUsingExtensionQueryis not enabled, so no extended queries are generated to match implicit compliance needs.
How to Verify Correct Configuration
- Confirm that the App/Admin server has correctly configured AIPROXYAPIENDPOINT (AI Proxy service root address, format example: http://aiproxy:3000, do not append paths such as /v1) and AIPROXYAPITOKEN (AI Proxy administrator Token, consistent with server configuration and non-empty). Do not fill in these two variables in workflow global variables, application API request headers, or model management pages.
- View retrieval logs to confirm that only content from datasets bound to the current project is returned. Verify that the
collectionIdsconfiguration takes effect. - Input a test question that includes sample specifications. Check whether the matching degree of returned results falls within the preset range. Verify the rationality of the
similarityconfiguration. - Compare recalled results with and without
datasetSearchUsingExtensionQueryenabled. Confirm that the extended query function works properly. - Check whether the sorted results prioritize matching core inquiry fields. Verify the effect of the
usingReRankconfiguration.
The values provided here are common starting points and should be evaluated against the user’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.