deepsight cloud · Anonymization Detect and mask personal information in texts.
Our own anonymization works without a language model: personal information is masked before the analyses run. Check the results before you share sensitive content.
A name, contact details and an indirect reference
“I complained to [PERSON] in HR, and you can reach me at [PHONE]. As the only student worker in the Lingen warehouse, I feel overlooked.”
A role and a location together can make a person identifiable, even without a name. The anonymization does not detect indirect references like these automatically – you check them yourself and mask them manually if needed.
Select an item to see the passage.
Illustrative example. The anonymization does not detect every piece of information – check results before sharing them. Masking does not replace a data protection assessment of your project. More on the limits of anonymization on our blog
Our own detection, your review
No language model
Detection is our own solution and works without a language model. Masking happens before the analyses run. If translation is switched on, texts are translated before masking.
Selectable categories
You decide which kinds of information are masked, such as email addresses or phone numbers. The names category is always switched on.
Manual masking
You mask anything that was missed by hand. Changes are versioned and can be undone.
Questions about anonymization
Personal information such as names, email addresses and phone numbers. It does not detect every piece of information, though: internal numbers such as customer or contract numbers are not a category of their own, and it does not detect indirect references to people automatically. You mask passages like these manually if needed.
Yes. Placeholders such as [PERSON] or [PHONE] replace only the detected information; the rest of the text stays intact. The analyses run on the masked text.
Texts are translated first and masked afterwards, because the anonymization supports fewer languages. This means texts are sent to the translation unmasked; the analyses run on the masked text. For sensitive data, check whether you need translation.
No. The AI follow-up during fieldwork and analyses started directly from a survey run without anonymization. To mask personal information, run a separate analysis of the survey dataset with anonymization enabled; this run counts toward your text allowance again.
That depends on the individual case, for example on indirect references in the text that the anonymization does not detect. Processing in deepsight cloud is governed by a data processing agreement (DPA) under Art. 28 GDPR.
Names and references to individuals can appear in comments from employee surveys, for example.
More features · Prepare data
See the feature with your own data
We will walk you through the analysis using an example from your field.