AI detection in the FOI workflow
Step 1
Document ingest
PDF from FOI case is analysed for entities and patterns.
Step 2
Generate proposals
AI flags names, IDs, contact details, and sensitive passages.
Step 3
Reviewer validation
Authorised staff accept, adjust, or reject each proposal.
Step 4
Logging and export
Decisions are logged for accountability and disclosure.
FAQ – AI detection
Does AI replace the reviewer?
No. AI proposes; your organisation makes the final redaction decision.
Is our data used to train models?
Contact us about processing and retention; default is EU hosting with limited retention.
Does it work on scanned PDFs?
Yes, with OCR for scans; native PDFs are faster.
自然语言处理
先进的NLP算法理解上下文，即使在复杂句子结构中也能检测敏感信息。
模式识别
跨格式识别模式：身份证号、电子邮件地址、电话号码等。
持续学习
系统持续从新数据中学习，随时间自动提高准确性。
机器学习脱敏工作原理
1
文档数据训练
我们的ML模型在大型文档数据集上训练，针对个人数据、身份证号和政府文档进行优化。
2
上下文分析
与简单模式匹配不同，我们的ML技术理解上下文。它区分真实个人数据和示例数据。
3
自适应改进
系统适应新的文档类型和格式。反馈循环确保准确性持续改进。