MIMS Final Project 2023

Privacy Assistant

Users are often unaware of how much they may be inadvertently revealing about themselves when writing social media posts (e.g. Twitter, LinkedIn), composing an email, or typing up a document. Privacy Assistant works to address this problem; it is a consumer-facing, intelligent writing assistant – like Grammarly, but for privacy! Using concepts around natural language processing, generalization, privacy engineering, and UX privacy by design principles, Privacy Assistant notifies the author of potential disclosures of personally identifiable information in their text and provides alternative suggestions to still convey useful information.

What is extra special about Privacy Assistant is that it is context aware. What may be private information will vary from one situation or user to another. There are moments when users find it necessary to disclose a specific piece of information but want to conceal other sensitive words surrounding it. Moreover, written text that communicates a similar message or situation can expose information that is more detrimental to one person than another. The suggestions that the tool provides users vary in levels of privacy and it even affords them the ability to opt out of the suggestions that do not suit their needs.

Our project’s goals

  • Provide a new way of categorizing and detecting PII or PII-adjacent words in written formats.

  • Add to the conversation about natural-language processing and contextual awareness through a privacy lens.

  • Develop an approach to applying privacy techniques to data in text formats.

  • Create a demo of the text assistant in action.

Last updated:

May 2, 2023