Manage recognition algorithms
Algorithms, rules, and tasks produce candidate classifications; accountable reviewers confirm and publish them.
Choose an algorithm
| Type | Use | Management focus |
|---|---|---|
| Algorithm library | Text, column, pattern recognition | Folders, expressions, samples, descriptions |
| Lineage cascade | Propagate candidates from confirmed objects | Lineage quality and scope |
| Machine learning | Model-based suggestions | Applicability and review sampling |
Availability depends on deployment/license. Check features and rights if absent.
Create an algorithm
- Open Data Security > Recognition Algorithms > Algorithm Library.
- Create a clear-purpose folder.
- Select New Algorithm and enter name, code, description, and applicable types.
- Configure expressions/parameters and test matches/nonmatches.
- Verify status, folder, and rule availability after saving.
Use synthetic samples, including nulls, boundaries, Chinese characters, and invalid formats.
Import, export, and maintain
Use system templates, unique codes, existing folders, and valid parameters. Inspect failed rows; task creation does not prove every algorithm is effective. Exports support review/backup, not version history. Resolve rule dependencies or create replacements before retirement/deletion and monitor subsequent results.
Check before publication
- Names/codes match business meaning.
- Positive, negative, and null tests pass.
- Results explain evidence and proposed classification.
- Rules reference intended algorithms.
- Task scope and owner are clear.
Next: Rules and tasks.