What data looks like for this category
Text annotation project data originates from customer-provided corpora, annotation results generated by internal annotation platforms, custom vocabularies, and quality inspection sampling results. Daily circulating documents include annotation specifications, sample packages, quality inspection reports, and issue feedback forms. Data updates occur daily or per annotation batch. Involved fields include text content, annotation tags, quality inspection results, annotator IDs, and timestamps. This data supports text annotation needs such as semantic entity recognition and intent classification, and differs from image annotation projects which focus on visual elements.
Constraints on permissions and data isolation workflows
Customer corpora count as sensitive business data. The batch-based update feature requires isolation rules to support permission verification for bulk data. Documents such as annotation specifications and quality inspection reports cover internal team operating standards and quality requirements, so access must be restricted for unauthorized personnel. Annotation fields include annotator information and quality inspection results, so only authorized roles may modify or view relevant fields during data circulation. The division of responsibilities between annotation teams and NLP algorithm teams also requires data isolation to separate annotation permissions, quality inspection permissions, and algorithm call permissions. This prevents cross-role unauthorized access to annotation samples and results.
How to configure settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
teamId | Bind to the exclusive team ID of the annotation project | Ensure only project members can access corresponding annotation data, isolate sensitive corpora across projects |
tmbId | Bind to the exclusive template ID of the annotation template | Restrict only roles configured with this template from modifying annotation specifications, prevent unauthorized changes to annotation rules |
| Role permission groups | Divided into three categories: annotators, quality inspectors, algorithm engineers | Match the responsibility boundaries of the annotation team. Only quality inspectors can view quality inspection reports, and algorithm engineers can only call desensitized sample data |
| Data access scope | Isolate units divided by annotation batch | Adapt to the batch-based update data feature, only personnel participating in the corresponding batch can access samples from that batch |
| Log retention period | Set based on actual measurement (usually 30–90 days) | Meet log tracing requirements for compliance reviews, while avoiding invalid storage occupation |
| Sensitive field desensitization | Perform desensitization processing on fields such as customer identifiers and original corpora | Protect sensitive information of customer business data, comply with compliance review requirements |
| Single sample access limit | No more than 100 entries | Limit the sample volume of a single request, avoid leakage of bulk sensitive data |
Three common configuration errors
- Symptom: After a team administrator sends a member invitation, the invited member does not receive any permission notification. Cause: The team permission notification switch is not enabled, so isolation rules cannot be synchronized to the invited member, preventing normal project access.
- Symptom: Annotators from other projects attempt to access sample packages from a different annotation batch, and receive a 403 permission denied error. Cause: The
teamIdparameter is not correctly bound, so permission verification does not match the authorized scope of the target project. - Symptom: When algorithm engineers call annotation samples, they retrieve undesensitized customer original contact information fields. Cause: Sensitive field desensitization rules are not configured, leading to missing sensitive information protection during compliance reviews.
How to confirm correct configuration
- Log in to the FastGPT v4.17.0 project management backend, enter the permission settings page of the target text annotation project, and confirm that
teamIdandtmbIdare correctly bound to the current project. - Invite a test member to join the project, verify whether the member receives a permission application notification to confirm the notification switch functions properly.
- Use an account with an unauthorized role to attempt accessing annotation specification documents, confirm that the documents cannot be viewed or modified.
- Trigger an annotation sample access request, check whether the system log includes corresponding permission verification records and desensitized sample data.
Note: The provided configuration values are common starting points. They should be assessed against the specific samples and operational scenarios of the respective deployment.
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.